Mihail Glovinsky
Offer Mihail work on your next project.
Private message 
Offer work 
Kyiv,
Ukraine
1 hour 36 minutes back
available for hire
11 Safes completed
4 months 20 days back
9 clients
on the service 7 years
Rating
Language proficiency level
Українська: fluent
Русский: fluent
English: intermediate
CV
No-Code Automation Developer | Make.com | Zapier | Airtable
Why Me?
Hi! I'm Mykhailo, your strategic automation partner. If you're looking for a way to eliminate repetitive tasks, optimize processes, and seamlessly integrate AI, you've come to the right place.
Why choose me over "just another freelancer"?
- Experience + Business Acumen: I've been in No-Code automation since 2018. Between mid-2019 and late 2024, I engaged in entrepreneurial activities, gaining an invaluable firsthand understanding of business processes, client pain points, and the critical importance of efficiency from a business owner's perspective. Now, I've returned to automation, blending this practical experience with cutting-edge AI technologies.
- Results, Not Just "Setups": My goal is measurable improvement for your business. Reducing manual work by 80%? Speeding up application processing by 50%? Increasing customer engagement by 35%? These aren't just numbers; they are real results achieved for my clients (see examples below).
- Specialization in No-Code & AI: I master key tools (Make, Manychat, Airtable) and skillfully integrate AI capabilities (OpenAI, Gemini, Voiceflow) to create powerful and flexible solutions without the need for complex development.
- Partnership, Not Formality: I dive deep into your challenges, ask the crucial (sometimes tough, but important) questions, propose optimal strategies, and work towards our shared success. I believe you value this approach too, right?
Highlighted Case Studies:
- For a Fitness Studio: Reduced administrative load by 60% thanks to an AI bot and integrations.
- For a Private Accountant: Cut down manual reminder work by 80%.
- For a Real Estate Company: Accelerated application processing speed by 50% using AI and data collection automation.
My Experience: Niches I've Worked In
I've had the pleasure of working with various businesses. Perhaps your business type is on this list?
- Education: Language schools, training companies, online courses...
- E-commerce: Online stores...
- Service Industry: Fitness studios, accountants, real estate agencies...
- Food Service (HoReCa): Street food businesses...
- Content Creators & Marketing: Automating content creation and publishing...
- Personal Finance & Productivity: Developing personal trackers...
- Psychology & Self-Development: Creating platforms for workshops/marathons...
My Services: What can I do for you?
Think some routine task can't be automated? Let's challenge that! 😉 Here are the main ways I can help:
- No-Code Automation Development & Implementation: (Analysis -> Tool Selection -> Workflow Building -> Testing -> Support)
- Smart AI Chatbot Creation (ManyChat, Voiceflow, Custom): (Goal Definition -> Logic Design -> Bot Development -> Integrations -> Analytics)
- Service & CRM Integration: (System Analysis -> Connection Setup -> Data Sync -> Unified Ecosystem)
- Marketing & Sales Automation: (Funnels -> Email Sequences -> SMM Automation -> Quizzes -> Ad Integration)
- AI Solution Implementation for Business: (Use Case Analysis -> Content Generation -> Transcription -> Personalization -> AI Assistants)
My Skills & Tools:
Core Tools (Expert Proficiency):
- Make.com: Building complex automation scenarios, integrating anything with anything via API/Webhooks.
- ManyChat: Developing advanced chatbots for Messenger, Instagram, WhatsApp, Telegram; automating sales funnels and support.
- Airtable: Creating flexible databases, CRM systems, backends for apps and automations.
Familiar Tools (Confident User in Projects):
- Automation Platforms: Zapier.
- No-Code/Low-Code: Google Sheets, Notion.
- Chatbot Development: Voiceflow, Telegram Bot API.
- Artificial Intelligence (AI): OpenAI (ChatGPT, GPT API, Whisper), Google Gemini, Claude, Perplexity, Dumpling AI, Runway AI, ElevenLabs.
- Design/Mapping Tools: Miro, Whimsical.
- Popular Service Integrations: Fondy, Calendly, Typeform, Mindbody, HubSpot, Zoom, Google Drive, Gmail, Trello, Slack, and others.
Education & Certificates:
- Course "AI Expert" (MASC Marketing Automation School, 2024-2025) - AI in business, automation, No-Code.
- Course "Chatbot Master" (Zushi Technology, 2018) - In-depth ManyChat training.
- Google Sheets Course (Genius Space, 2022) - Data analysis and reporting.
- Getting Started with ChatGPT (Prometheus, 2024)
- English B1 (Campster, 2022).
Don't put off until tomorrow what you can automate today! Write to me, tell me about your challenge, and together we'll find the best No-Code and AI solution. Ready to discuss your project?
Tags / Keywords:
Make.com, Zapier, Integromat, No-Code, Low-Code, Business Automation, Business Process Automation, AI Automation, Artificial Intelligence, OpenAI, ChatGPT, Gemini, ManyChat, Chatbots, Messenger Bot, Instagram Bot, WhatsApp Bot, Telegram Bot, Airtable, Google Sheets, CRM Automation, API Integration, Webhooks, Voiceflow, Voice Bots, Marketing Automation, Auto Funnels, Sales Funnels, Lead Generation, Transcription, Content Generation, Workflow Automation, Process Optimization, Ukraine, Ukrainian Freelancer.
Skills and abilities
Programming
-
AI & Machine Learning
from 27 USD for hour
-
Bot Development
from 27 USD for hour
-
Data Parsing
from 27 USD for hour
Design & art
-
AI Art
from 27 USD for hour
Services
-
AI Consulting
from 27 USD for hour
-
Customer Support
from 27 USD for hour
Photo, Audio & Video
-
AI Speech & Audio Generation
from 27 USD for hour
-
Video Creation by Artificial Intelligence
from 27 USD for hour
Writing
-
AI Content Creation
from 27 USD for hour
Portfolio
- All works 33
- System & Network Administration 1
- Databases & SQL 2
- Email Marketing 1
- Accounting Services 1
- Data Parsing 1
- Information Gathering 2
- Customer Support 1
- AI & Machine Learning 15
- Bot Development 6
- AI Consulting 1
- Video Creation by Artificial Intelligence 1
-
Custom AI models — from training to deployment on your own server
Learn to create your own AI models for specific tasks and tones — so that the client can have their own AI, rather than relying solely on general services like ChatGPT. The final product is a set of working, compact models that operate on a private server alongside automation, without paying external providers for each message.
Key requirements:
… Zero costs: the entire process using 100% free tools, without paid cloud services.
Fair comparison: test three model sizes and find the best balance of quality and price.
Ease of use: models must be small enough to run even on a weak old laptop.
Integration readiness: models in a format that connects directly to n8n automations.
Stability: bypass technical limitations of free training platforms.
How it works:
The idea was to be able to offer the client their own trained model — not just a connection to someone else's service — and to do this entirely with free tools.
I trained three models and compared them fairly: I trained three models of different sizes on the same data and tested them on the same questions to see which provided the best answers for the money. This gives a clear basis to advise the client on the right model for their budget.
I solved the complex limitation of the platform: free training services have a known issue that breaks modern models. I found a reliable workaround that allowed me to train a model where standard instructions simply do not work.
I selected the right model size for the limitations: the largest model would have taken over a day to train and would not fit within the free time limit. I switched to a smaller multilingual model — strong in Ukrainian — and reduced training to a few hours without any real loss of quality.
I built a reproducible production process: training → packaging → compression → deployment. This same process is now used to train an AI assistant that works directly with n8n automations.
I proved that data quality is everything: I analyzed several public datasets and showed that weak responses come from poor training data, not from the model itself — a key takeaway for any future custom AI project.
Result:
A working set of custom AI models and a process that can be reapplied to real client projects:
A ready-to-use AI model published online (compact version ~1 GB), which works even on an old laptop.
A reproducible process for training, packaging, and deploying custom models — suitable for future client projects.
A fair comparison of "quality vs price" across three model sizes — a basis for recommending the right model for the budget.
Production readiness: models are deployed on a private server and connect directly to n8n automations via a standard API.
#n8n #AI #LLM #FineTuning #CustomAI #SelfHosted #MachineLearning #Automation #NoCode #HuggingFace #Ollama
-
226 USD Automated lead generation and tender monitoring
AI & Machine Learning
Goal:
Create an autonomous lead generation system that daily scans the Prozorro registry for two categories of opportunities: concluded contracts (warm leads — organizations that are already purchasing relevant products) and active tenders (where proposals can be submitted). The final product is a structured CRM database in Airtable with complete contact information of the client, participants, and proposal prices.
Key Requirements:
…
Two parallel workflows: isolated processing of contracts (/contracts) and active tenders (/tenders?status=active.tendering)
Extended filtering: 5 CPV codes + 15 marker brands + minimum amount of 50,000 UAH
Cursor pagination with historical backfill: scanning Prozorro from 01.01.2025 without losing position between launches
Detailed CRM structure: contacts (EDRPOU, email, phone), auction participants with prices, links to the proposal registry
Deduplication at the Airtable level: preventing duplicate entries during continuous operation
Hybrid logic: closed tenders that pass the filter automatically enrich the lead database with participant data for future email campaigns
My Contribution:
The client trades energy equipment (batteries, UPS, solar panels) and spent hours daily on manual monitoring of Prozorro. The task was not just to "send tenders," but to build a sales-ready lead conveyor with a complete market breakdown.
Architectural decomposition: instead of one "thick" flow, I divided the system into two independent workflows — for contracts and active tenders. Each has its own cursor, launch schedule, and Airtable table. This allowed scaling sources without mutual blocking.
Cursor pagination with self-healing logic: Prozorro returns a wrapping cursor in the format {timestamp}.{seq}.{hash}. I implemented a protective layer that validates the cursor before use and automatically restores it from a valid date if the API returns an anomaly (for example, a reset to 2015). Without admin intervention.
Two-tier data enrichment: for leads, I created a cascade of HTTP requests — first the contract (buyer, supplier, amount), then the tender (participants, bids, proposal registry). A separate JavaScript node combines these streams and forms JSON with an array of participants (name, EDRPOU, email, initial + final bid), ready for integration with email campaigns.
Deduplication through Airtable Search: before each entry, the flow searches by contractID/tenderID, allowing the workflow to be safely restarted as many times as needed without creating duplicates.
Routing of closed tenders: tenders that ended before processing are not discarded — IF logic redirects them to the "Leads" table with a complete list of participants. This way, the client receives not only potential buying customers but also competitor suppliers for market analysis.
Result:
A functioning system that daily enriches the client's CRM without any human intervention:
2 data sources in one database: Airtable with two tables ("Leads" and "Active Tenders"), each with contacts, prices, documents, and links
Readiness for email marketing: structured data of participants (with emails and phones) can be directly sent to Mailchimp/SendPulse/own script
Saving 2-3 hours/day: instead of manual googling, the client immediately receives a table with 18 fields
Historical coverage: backfill from 01.01.2025 allowed the formation of a warm lead database over 16+ months even before launch
Stability 24/7: the system runs 4 times a day on schedule, with retry logic for API errors and self-recovery of the cursor
#n8n #Airtable #LeadGeneration #SalesAutomation #Prozorro #API #WorkflowAutomation #JavaScript #CRM #DataPipeline #BusinessAutomation #NoCode #B2B #LeadGeneration #Automation
-
150 USD Automated monitoring of daily sales of Prom and Rozetka
AI & Machine Learning
Goal: Automate the collection and consolidation of daily financial analytics from various marketplaces (Prom, Umall, Rozetka) through CRM SalesDrive in Telegram. Key challenge: to merge data from disparate sources (different formId) and ensure the accuracy of profit calculations and "sales hits" amid unstable transmission of system dates via API.
Solution: Multi-threaded architecture on n8n (self-hosted on Railway):
… 1. Data Extraction (SalesDrive API Pipeline)
Multi-Source Fetching: Implemented separate HTTP requests for the PROM + UMALL branches and ROZETKA. Specific authentication headers were used for each source.
Precision Filtering: Configured strict filtering by time windows (00:00:00 – 23:58:59) through orderTime and createTime parameters, eliminating the pulling of "historical" orders when edited by managers.
Status Validation: The system automatically filters out technical junk and canceled requests (status 32/51), ensuring that only real "cash" enters the report.
2. Analytics Engine (JS Code Logic)
Financial Aggregator: Code Node acts as a backend processor: it parses arrays of orders, converts string data to numbers (parseFloat), and sums up turnover and net profit in real-time.
Product Hit Detection: The algorithm dynamically counts the number of units sold for each SKU, identifies the leader of the day ("Sales Hit"), and outputs its name along with the number of sales.
Data Normalization: Automatic unification of product names and SKUs, even if they are missing in individual objects of the products array.
Result:
Reporting Design: Formation of a strict business report in Telegram using a monospaced font for easy reading of numbers and instant copying of EDRPOU/amounts with one tap.
Operational Control: The owner receives a complete picture of sales in 1 second without the need to log into CRM and manually build filters.
Reliability: Thanks to hosting on Railway and optimized connection timeouts to the database (PostgreSQL), the system reliably processes requests even during peak load hours on marketplaces.
#n8n #SalesDrive #ECommerce #Prom #Rozetka #BusinessIntelligence #Automation #API #Railway
-
135 USD Synchronization architecture: Notion → Reclaim.ai
AI & Machine Learning
Goal: Automate the scheduling of production tasks from Notion to Reclaim.ai via Google Tasks. Key challenge: implement reliable deduplication without changing statuses in Notion and bypass strict Google API quotas.
Solution: Two-level architecture on n8n (Railway):
… 1. Collection and Validation (Notion Pipeline)
Smart Deduplication: The system ignores duplicate Notion triggers by matching page IDs with its own database (Data Table). This allows the status PRODUCE to remain static.
Dynamic Time-Window: Filtering tasks by a 14-day window (REZERWACJA), which excludes scheduling of archived records.
JS Hours Parser: Code Node automatically converts free input of hours into the format (duration: Xh), understandable for Reclaim.ai AI algorithms.
2. Delivery and Optimization (Queue Engine)
Quota Management: Implemented batching and retries, which eliminated 403 Quota Exceeded errors during bulk operations with Google API.
Asynchronous Flow: Distribution into "Collector" and "Sender" through a status queue (PENDING -> SENT), ensuring 100% delivery of each task.
Result:
Sync Speed: Appearance of the task in the calendar within 1–5 minutes.
Stability: Complete automation without "manual" support of statuses in Notion.
Scalability: Ready infrastructure for scaling to other departments of the company.
#n8n #Notion #ReclaimAI #Automation #Backend #API
-
902 USD AI lead identifier: Bitrix24 + Binotel
AI & Machine Learning
Objective: Develop an intelligent CRM marketing automation system for identifying anonymous leads in Bitrix24. Requirement: ensure automatic customer recognition through the analysis of incoming calls from Binotel telephony, minimize costs for AI analysis through batch deduplication, and clean the database of "junk" contacts.
My Contribution / Solution: Designed a multi-level architecture on self-hosted n8n, integrating Bitrix24 API with models (GPT-4o). Implemented logic for maintaining data context during complex workflow branches.
… 1. Intelligent Media Engine (Analysis and Identification):
Multi-Step AI Transcription & Analysis: Implemented a system for extracting audio recordings from Bitrix Activity entities. Used neural networks for transcription and semantic analysis of dialogues to identify customer names, company names, and types of requests.
High-Precision Filtering: Implemented strict filtering of the incoming stream: ignoring outgoing calls (DIRECTION: 1), cutting off conversations shorter than 40 seconds, and spam detection. This allowed focusing AI resources only on targeted incoming leads.
2. Batch Processing and Data Integrity (Optimization):
Batch Deduplication Standard: Developed a mechanism for comparing incoming data with the existing database based on the principle [Input] - [DB] = [New]. This eliminated the reprocessing of archived calls (2024–2026) and reduced costs for neural network API.
Source of Truth Recovery (V16): Resolved the issue of context loss (Activity ID, Phone) during successful API requests to Bitrix24. Created an architecture where the final node Normalize Data refers to the initial state of the iterator (Process Calls3), ensuring 100% field completion in final logging.
3. Reliability and Infrastructure Management:
Optimized the operation of the self-hosted n8n instance for bulk processing of large archives. Implemented a data cleaning strategy, disabled logging of successful runs to save disk space, and implemented automatic database compression.
Archival & Real-time Hybrid: The system is configured for hybrid mode: deep processing of historical archives (depth up to 2 years) and daily monitoring of new contacts "on hot trails."
Result: Created an autonomous backend product for automatic enrichment of CRM data:
Data Enrichment: Automated the identification of over 80% of anonymous incoming calls, transforming "Phone call from..." into named contacts with a request history.
#n8n #Bitrix24 #Binotel #AIAutomation #GPT4 #Backend #CRMIntegration #NoCode #DataEngineering
-
300 USD Multimodal platform based on n8n
AI & Machine Learning
Goal: Create a professional tool for generating high-quality AI video content (Cinematic Trailers) with a hybrid pricing model. Requirement: ensure quality at the Kling/Luma level at optimal cost, implement a balance management system, and automate post-production (branding).
My Contribution / Solution: Developed a multi-level architecture on n8n, which integrates advanced SOTA models (Kling, Veo, Luma) through the FAL.ai API. Implemented dynamic model selection logic based on user budget and technical video requirements (duration, frame rate).
… 1. Hybrid Media Engine (Quality and Cost Optimization):
Multi-Model Orchestration: Implemented logic for switching between models: Kling 1.0 (for budget generations ~$0.45), Veo 3.1 (price/quality balance ~$1.2), and Kling 2.1 Master (VIP quality). This allowed the client to offer flexible rates for different user segments.
Prompt Engineering & Validation: Implemented a system for processing complex cinematic prompts. Automated parameter checks (duration, aspect ratio) at the input stage, preventing API errors (e.g., validating duration 4s/8s for the Veo model) and saving costs on failed requests.
2. State and Finance Management System (CRM & Balance):
Atomic Credit System: Designed an "internal currency" (balance) system. Each generation deducts funds only after successful confirmation from the API, eliminating user money loss during neural network technical failures.
Concurrency & Request Tracking: Resolved the issue of record duplication during long operations (video generation takes 5-10 minutes). Used a unique Transaction ID identification mechanism, preventing repeated node launches during repeated callback requests from Telegram.
3. Photo-to-Video Workflow:
Dual-Path Generation: Implemented two operational scenarios: video generation based on the user's uploaded photo (Custom Input) and video creation from pre-generated AI images (Full-AI Loop).
Result: Created a scalable SaaS product for AI video generation, ready for commercial use:
Cost Control: The cost of generating a 10-second video has been reduced to $0.45 while maintaining high quality.
High Reliability: The system reliably processes long rendering queues (async processing) without "hanging" the workflow.
Monetization Ready: Fully developed logic for CRM, balances, and access levels (Basic/VIP).
#n8n #GenerativeAI #KlingAI #VideoAutomation #Backend #APIIntegration #SaaSDevelopment #LowCode
-
271 USD Intelligent order management system
Bot Development
Goal: Develop a fault-tolerant order acceptance system (E-commerce Bot) with logic to protect against data collisions. Requirement: the system must operate as a State Machine (finite automaton), dynamically adjusting the interface based on the availability of products in the database and ensuring the "cleanliness" of input data even before the processing stage by the manager.
My Contribution / Solution: Designed and implemented a "State-Aware Architecture" based on n8n, which manages the user lifecycle from entry to deal finalization. A Low-Code + Custom JS approach was used to bypass the limitations of standard nodes.
… 1. "Gatekeeper" Architecture (Session Management):
Logic Engine & Concurrency Control: A strict "Face Control" algorithm (Switch Node Logic) was implemented. The system checks the database for any unfinished transactions (statuses "Pending", "Payment Wait") before starting a new process. This prevents the creation of duplicates and "junk" records.
Session Reset & Cleanup: A mechanism for forcibly resetting stuck sessions (force_cancel) was implemented, allowing the user to resolve state conflicts independently without contacting support.
2. Dynamic Frontend (Telegram UI):
JSON-Generated Interface: Instead of hardcoding buttons, dynamic menu generation was implemented (JavaScript Code Node). The bot queries the warehouse database (n8n Table/Airtable), retrieves the current product catalog, and renders the keyboard "on the fly." This allows adding products (/add) without restarting the bot.
Raw HTTP Requests: To bypass the limitations of standard n8n nodes and eliminate UI glitches, direct POST requests to the Telegram API were used. This ensured stable operation of complex reply_markup objects and correct transmission of callback data.
3. Data Integrity & Validation:
Smart Parsing: A combination of regular expressions (RegEx) and conditional operators for input validation (for example, distinguishing the amount "200" from the phone "050...").
Database Locking: The use of atomic updates of rows by unique IDs (instead of ChatID), which solved the problem of data overwriting when multiple operators or instances were working simultaneously.
Result: An autonomous sales system was created that does not require technical supervision:
Zero-Conflict Database: The number of duplicated or erroneous orders has been reduced to 0 thanks to state-management logic.
Scalability: Adding new product items takes seconds through admin commands, automatically updating the interface for all users.
User Experience: The system itself "guides" the client, blocking illogical actions and offering contextual scenarios for error recovery.
#n8n #JavaScript #BackendArchitecture #StateManagement #TelegramBotAPI #ECommerceAutomation #ErrorHandling #DatabaseDesign
-
451 USD Telecom traffic analysis system with AI-Vision
AI & Machine Learning
Objective: Automate the critical business process of a telecom company — analyzing incoming rate sheets from VoIP traffic suppliers. The problem was the variety of formats: suppliers send prices in Excel, CSV, PDF, and even images (screenshots in messengers). Manual processing took hours, leading to missed profitable deals in a dynamic market. Requirement: The system must be "omnivorous," identifying profitable (BUY) and unprofitable (SELL) directions by comparing them with the internal market API and instantly notifying managers.
My Contribution / Solution: The solution was implemented on Self-hosted n8n, using Google Drive, OpenAI (GPT-4o / Turbo), and Google Sheets. The architecture is built on a "Parent-Child" principle for scalability and fault tolerance.
… 1. Workflow "Omni-Channel Ingestion" (Parent Processes):
Routing and Queue: Implemented Smart Queue logic. The system scans Google Drive, identifies the file type (.xlsx, .csv, .pdf, .png), and processes them one by one (Batch Size: 1) with a 20-minute interval to avoid API overload and limits.
AI Vision & OCR: A cascade was created for processing "unreadable" formats (PDF/Images):
CloudConvert: Conversion of multi-page PDFs into high-quality PNGs (300 DPI).
GPT-4o Vision: Utilization of a multimodal model for visual reading of tables from images, where regular parsers are powerless.
Smart CSV Parsing: For large text files, a "Chunking" algorithm was implemented — breaking the text into packets of 40 lines for processing by a lighter GPT-4 Turbo model, significantly saving the client's budget.
2. Workflow "Analytical Core" (Child Workflow):
Data Enrichment: A complex matching algorithm (JavaScript) was implemented. The system normalizes country names (for example, correcting "Dr Congo" to the official name), determines MCC/MNC codes from the internal directory, and classifies traffic type (Direct/HQ/SS7/Sim) based on file metadata.
Market Intelligence: Integration with an external API (interconnect.solutions). Each line of the price is checked in real-time for market median.
Logic Engine: Automatic margin calculation. The system assigns a BUY status (if the price is below the market) or SELL, and sorts offers from the most profitable.
3. Reliability and UX:
Error Handling: A global error interceptor and local Retry strategies (3 attempts) for unstable HTTP requests were configured.
Reporting: The final result is formed as an interactive HTML report in Telegram with links to the original and processed file, as well as the Top-15 recommendations for managers.
Result: The client received a fully automated traffic procurement department:
Versatility: The system processes any incoming file, from Excel to screen photos.
Response Speed: The time from receiving the file to decision-making was reduced from hours to minutes.
Economic Effect: Managers receive ready "signals" (Buy Alerts) and do not waste time manually comparing thousands of lines.
Stability: Thanks to queues and request optimization, the system operates 24/7 without API failures.
#n8n #OpenAI #GPT4o #ComputerVision #Automation #Telecommunications #VoIP #JavaScript #GoogleDriveAPI #DataEngineering #CloudConvert #TelegramBot
-
338 USD Digital co-founder.
AI & Machine Learning
Goal: Create an intelligent AI partner for the owner of a construction and development company. The system was to combine an offline knowledge base (Obsidian) with the power of cloud AI (OpenAI GPT-4o). Key requirements:
- RAG (Retrieval-Augmented Generation): Responses must be based solely on the internal regulations and documents of the company.
… - Bidirectional communication: The agent must not only "read" the database but also "write" to it (create new files/regulations upon command in Telegram).
- Resource efficiency: Smart indexing to avoid re-reading unchanged files.
My Contribution / Solution:
The solution is built on a Self-hosted n8n (Railway), vector database Supabase, and Google Drive cloud storage. The architecture consists of 3 complex workflows:
1. Workflow "Smart Indexer" (ETL Pipeline):
Google Drive (Recursive Search): A complex algorithm for searching files (.md, .txt, .pdf) across the entire drive has been implemented, traversing nested folders and filtering out "foreign" files.
Incremental Sync (Cost Savings): Logic for comparing metadata has been developed. The workflow compares files from the drive with the file_tracker table in Supabase (SQL). Only new or modified files are sent for processing (Embedding). This saves up to 90% of OpenAI tokens.
Vectorization: Text is broken into chunks, converted into vectors (OpenAI Embeddings), and stored in Supabase.
2. Workflow "Brain" (Conversational AI Agent):
AI Agent (LangChain): Uses the GPT-4o model with a custom system prompt "Digital Co-Founder".
Long-term Memory: Connected to Postgres Chat Memory (in Supabase), allowing the bot to remember the context of dialogues indefinitely.
Vector Store Tool: A search tool has been implemented that uses a custom SQL function match_documents to find the most relevant answers in the knowledge base.
3. Workflow "Hands" (File Generator Tool):
Autonomous content creation: The agent can invoke this sub-workflow to create new documents.
Smart Parsing (JavaScript): A sanitizer script has been written that parses the AI response (even if it comes in a non-standard format) into filename and content.
Write-back: The file is uploaded to Google Drive, after which it is automatically synchronized with the client's local Obsidian via Google Drive Desktop.
Result:
The client received a fully autonomous knowledge management system:
"Live" Database: Any change in an Obsidian note automatically goes into the "brain" of the bot.
Strategic partner: The owner can consult the bot regarding strategy, and the bot responds based on the history and context of the company, rather than general phrases.
Routine automation: The bot works as a secretary — creating drafts of contracts, ideas, and plans directly in the owner's working folder.
Reliability: Issues with server timeouts and data duplicates have been resolved through SQL optimization and Railway settings.
#n8n #OpenAI #RAG #Supabase #VectorDatabase #PostgreSQL #Obsidian #KnowledgeManagement #WorkflowAutomation #JavaScript #Railway #SelfHosted #GoogleDriveAPI #AIagent
-
113 USD AI lead analyzer from 8 RSS feeds
AI & Machine Learning
Goal:
Create an autonomous AI assistant for monitoring and qualifying freelance projects for a marketing agency (SEO/SMM/PPC). The key requirement is the aggregation of data from 8+ different RSS feeds, complete automatic deduplication of projects to save costs, and intelligent analysis of each unique lead using OpenAI before sending it to the client's Telegram group.
… My Contribution:
The project had two fundamental problems:
Information Noise: Programming (Python, PHP) and design projects were landing in relevant marketing categories (e.g., "AI" or "Bots").
Mass Duplicates: The same project often appeared in 3-4 different RSS feeds simultaneously, leading to 3-4 identical notifications and, worst of all, 3-4 times the payment for analysis in OpenAI.
My contribution was in designing a complex, multi-stage architecture of a "pipeline" in n8n. I developed a "bulletproof" deduplication system that is the heart of this workflow. Instead of simple filtering, I combined "streaming" deduplication (within a single run) with "persistent memory" (n8n Data Tables), ensuring that no project is analyzed twice, regardless of when and where it came from.
Solution:
The final solution is a single n8n workflow that runs on a schedule every 10 minutes and consists of 5 logical blocks:
1. Collection and Aggregation Block:
The Schedule Trigger initiates 8 parallel RSS Read nodes, each monitoring its category (SEO, SMM, PPC, Leads, etc.).
The Merge (Combine All) node collects all 8 streams into one array of projects.
2. Preparation Block:
The Set (Edit Fields1) node standardizes the data and creates a fullText field (from title and content) for future analysis.
3. Deduplication Block (Key Stage):
Data Table (Get row(s)): Loads from "memory" (Processed_Leads) the complete list of guids of all previously processed projects.
Merge (Merge_Deduplicate): Uses the keepNonMatches mode. It compares the stream of new projects (Input 1) with the list of old guids (Input 2) and only passes on those projects that are not in "memory."
Remove Duplicates (Node 1): Removes duplicates within the current run (in case one project came from 2 RSS feeds simultaneously).
Remove Duplicates (Node 2): An additional "on-the-fly" check against n8n's internal memory, ensuring 100% uniqueness.
4. AI Analysis and Storage Block:
Message a model (OpenAI): Receives only unique projects. The GPT-4 prompt analyzes fullText and returns JSON with a score, reason, and "trash" marker.
Data Table (Insert row): Immediately records the guid of the just-analyzed project in "memory" (Processed_Leads), so it will never go through deduplication again.
5. Notification Block:
Code (JavaScript): A "sanitizer" node that cleans the title and reason of special characters (*, _, [ ]), which could break Telegram formatting.
Telegram (2 nodes): Sends a perfectly formatted, analyzed message with AI scoring to two recipients — me (for monitoring) and the client's working group.
Result:
A fully autonomous AI assistant has been created that monitors 8 sources 24/7. The client received a system that:
Guaranteed saves money: 100% of duplicates are filtered before sending to OpenAI, preventing unnecessary API costs.
Saves time: The client receives not a "raw" stream, but already analyzed leads with a score and a brief summary.
High relevance: The intelligent prompt in OpenAI further filters out "trash" (is_trash: true) that slipped through the RSS.
Reliability: Using Data Tables as persistent "memory" ensures that even when the workflow is restarted, the system does not send old projects.
#n8n #OpenAI #GPT4 #WorkflowAutomation #LeadGeneration #RSS #APIIntegration #DataTables #Deduplication #Telegram #JavaScript #Freelance #MarketingAutomation #SEO #PPC #SMM #Automation #LeadGeneration #Marketing
-
95 USD Development of the WayForPay payment gateway on n8n for a Telegram bot
AI & Machine Learning
Goal:
Create a fully autonomous subscription management system for a Telegram bot. The key requirement is integration with the Ukrainian payment service WayForPay for automatic invoicing and access activation after payment. The system must reliably track subscription statuses, update user data in Google Sheets, and meet the strict security requirements of the WayForPay API.
My Contribution:
… The project started with a fundamental technical problem: standard no-code platforms, such as Make.com, lack built-in tools for generating and validating HMAC-MD5 cryptographic signatures, which are mandatory for working with WayForPay. This made direct integration impossible.
My contribution was to develop a new architecture from scratch on self-hosted n8n, which completely addressed this issue. I designed a robust two-component system, separating the logic of invoice creation and payment processing into two distinct but closely integrated workflows, ensuring maximum stability and ease of debugging.
Solution:
The final solution consists of two optimized workflows in n8n that provide a complete payment processing cycle.
Workflow 1: Invoice Creation
At the heart of this process is a chain of Crypto and Code nodes.
Request Preparation: Using the Code node, a perfectly structured JSON request body is dynamically formed, ensuring the correctness of data types (numbers, arrays) required by the API.
Signature Generation: The Crypto node creates a unique HMAC-MD5 signature for the outgoing request.
Invoice Creation: The final HTTP Request sends the signed request to WayForPay, receiving a unique payment link (invoiceUrl) in response, which is immediately sent to the user in Telegram.
Workflow 2: Payment Processing and Validation
This workflow is triggered via Webhook after successful payment by the client.
Data Parsing: The first Code node proved critical for "unpacking" the data, as WayForPay sent the webhook in an unexpected x-www-form-urlencoded format.
Security Check: The Crypto -> IF chain performs the most crucial function — it recreates the HMAC-MD5 signature from the received data and compares it with the signature from WayForPay. The process continues only if there is a complete match and the status is Approved.
System Update: In case of successful validation, the Google Sheets node updates the user's subscription status, sets a new expiration date, and resets usage counters. The user receives an instant notification in Telegram.
Transaction Completion: The final chain of nodes generates another signature and sends the correct response to WayForPay via Respond to Webhook, confirming the successful receipt of the webhook.
Result:
Successfully developed and implemented an autonomous payment system that fully automated the subscription management process in the Telegram bot. The client received a reliable and secure workflow that operates 24/7 and guarantees:
Reliable integration with the complex API of the WayForPay payment service.
Complete security through cryptographic validation of each request.
Instant activation of subscriptions and user notifications without any manual intervention.
Stable operation due to the separated architecture and correct handling of responses.
#n8n #WayForPay #APIIntegration #WorkflowAutomation #Telegram #JavaScript #PaymentGateway #Webhook #HMAC #GoogleSheets #SubscriptionAutomation #BusinessAutomation #Automation #PaymentGateway
-
100 USD Smart lead distribution system for CRM based on n8n
Databases & SQL
Objective:
Create a fully autonomous system for distributing incoming leads between two different CRM systems (campaigns) based on complex business rules. Key requirements included: dynamic management of daily limits for each campaign, processing leads exclusively during designated working hours, sequential sending with random delays, and developing a reliable deduplication system to prevent duplicate contacts.
My Contribution:
… The project began with a challenge: the existing lead processing was inefficient, did not support quota distribution, and resulted in sending duplicates, which lowered the quality of managers' work.
My contribution involved a complete rethinking and development "from scratch" of a new, reliable architecture on self-hosted n8n. I transitioned from simple linear logic to a more advanced "batch processing" architecture, significantly improving the system's performance and reliability.
Solution:
The final solution is a single, optimized workflow in n8n, at the heart of which is a custom Code node in JavaScript. This "brain" of the system performs all analytical work in one pass:
Loads context: Makes a single efficient API request to Google Sheets to retrieve all historical information, avoiding API limit breaches.
Performs deduplication: Identifies and filters out new leads whose emails have already been successfully processed.
Distributes by limits: Dynamically assigns each unique lead to a campaign (Campaign A or Campaign B), tracking daily quotas.
After the analytical block, a loop (Loop Over Items) is activated, ensuring sequential, individual processing of each lead with a random delay before the final sending via HTTP Request.
Result:
Successfully developed and implemented an autonomous system that fully automated the lead distribution process. The client received a reliable workflow that operates 24/7 and guarantees:
- Clear adherence to daily quotas for each campaign.
- Complete elimination of duplicates.
- Optimal load on external service APIs.
#n8n #JavaScript #WorkflowAutomation #BusinessLogic #APIIntegration #GoogleSheets #CRM #Automation #NoCode #LeadDistribution #Debugging #WorkflowArchitecture #BusinessAutomation #Automation
-
95 USD Personal Telegram assistant for managing tasks in Trello
Bot Development
Goal:
Develop an intelligent Telegram assistant that completely takes over the routine task management in Trello. Key requirements: creating a "single window" in Telegram for quick task setting, receiving proactive reminders, and the ability to interact with the task manager using natural language, minimizing the need to open Trello.
My Contribution:
… The main challenge in this project is not the technical limitations of the API, but the human factor: constant "context switching" between the messenger, where communication takes place, and the task manager, which steals time and focus.
My contribution involved developing a comprehensive bidirectional system based on self-hosted n8n, which transforms Telegram into a full-fledged command center for Trello:
Intelligent processing of incoming commands: I developed a workflow that listens to messages in Telegram and, using a logical Switch node, instantly determines the user's intent:
Simple task: If the message is a standard task, it is immediately sent via the API to Trello to create a new card in the specified list. The user receives instant confirmation.
Complex request: If the command requires analysis or response generation (for example, "summarize my tasks for today"), the request is forwarded to the OpenAI model (ChatGPT) for processing, after which a structured response is sent to the user.
Proactive reminder system: A second, independent process was created that operates on a schedule (Schedule Trigger). It automatically:
Once a day connects to Trello and collects all cards from a specific list (for example, "Today").
Checks if there are tasks in this list.
If tasks are present, it generates a concise report and sends it to Telegram as a morning reminder. This ensures that no important task is missed.
Result:
Successfully launched a personal assistant that operates 24/7 and significantly enhances personal productivity. The solution completely eliminates the need to manually transfer tasks from chats to Trello and constantly check the board.
The client received:
Time savings: Dozens of routine operations are now performed with a single command in Telegram.
Increased focus: Reducing the number of switches between applications allows for better concentration on task execution.
Reliable control: Automatic daily summaries ensure that priorities are not forgotten.
The final architecture built on n8n is extremely flexible: new commands can be easily added, other services (such as Google Calendar) can be integrated, or the reminder logic can be changed without rebuilding the entire system.
#n8n #Trello #Telegram #TelegramBot #Automation #NoCode #WorkflowAutomation #API #APIIntegration #OpenAI #ChatGPT #Productivity #TaskManagement #Automation #ChatBot
-
90 USD Multi-channel tender monitoring system
AI & Machine Learning
Goal:
Create a single, fully automated notification flow in Telegram that aggregates relevant tenders and projects from two completely different sources: the Ukrainian state portal Prozorro and the international API of the World Bank.
Key Requirements:
…
Parallel Processing: Simultaneous monitoring of national and international sources.
Data Transformation: Converting complex and heterogeneous data into a single, standardized format.
Instant Notifications: Prompt delivery of formatted reports in Telegram.
Scalability: Architecture ready for easy integration of new APIs in the future.
My Contribution:
The project started with a challenge: information about relevant tenders was scattered across isolated state and international portals, each with a unique API structure, data format, and access rules. Direct aggregation was impossible without comprehensive processing.
My contribution was in designing and developing "from scratch" a unified architecture on self-hosted n8n that combined these heterogeneous data streams into one powerful tool.
Analysis and Integration of Heterogeneous APIs: I conducted an in-depth analysis of two completely different APIs — Prozorro and World Bank. This included studying documentation, identifying the correct endpoints, query parameters, and, most importantly, the structure of their responses.
Transformation of Complex Data: The World Bank API returned data in an extremely atypical structure (an object of objects instead of an array). To address this issue, I wrote a custom script in JavaScript in the Code node that parsed this structure, normalized it, and transformed it into a standardized format ready for further processing.
Building a Parallel Workflow: I developed a single workflow that runs on a schedule and executes two parallel branches for each source. The system manages the full cycle:
Automatic retrieval of lists of tenders and projects.
Iterative processing: Obtaining detailed information for each record separately.
Dynamic formatting of data into readable messages using the Code node.
Merging streams through the Merge node to create a single notification queue.
Result:
A fully autonomous "radar" for tracking tenders has been created, operating 24/7.
Single Information Channel: The client receives notifications from Ukrainian and international sources in one Telegram chat.
Time Savings: The system fully automates the manual process of monitoring multiple websites.
High Scalability: The architecture with parallel branches and the Merge node allows for easy addition of new sources (e.g., TED, UNGM) without the need to rebuild the entire logic.
Reliability: The solution runs on its own instance of n8n, ensuring complete control, security, and absence of third-party limitations.
#n8n #API #APIIntegration #Automation #Prozorro #WorldBank #Telegram #TelegramBot #NoCode #JavaScript #WorkflowAutomation #DataParsing #BusinessAutomation #Automation #ChatBot
-
564 USD Project: AI assistant for analyzing team meetings
AI & Machine Learning
Goal:
Create a comprehensive AI system for the automatic processing of audio recordings from team meetings. Key requirements include transforming unstructured conversations into structured, analysis-ready data; creating multi-level summaries for different roles (Owner, Team Lead, BizDev); and developing an interactive AI assistant for instant access to the corporate knowledge base via Telegram.
My Contribution:
… The project started with a challenge: key company information — decisions, problems, and tasks — was "locked" inside hour-long audio files. This made searching and analyzing it practically impossible, forcing employees to waste time re-listening.
My contribution involved designing and developing "from scratch" a cohesive, multi-component architecture on self-hosted n8n, which transformed passive audio recordings into an active and intelligent resource.
I strategically chose the stack PostgreSQL + Supabase, which allowed combining the reliability of a relational database for structured reports with the power of a vector store for semantic AI search.
The system consists of three interconnected workflows that ensure the complete data lifecycle:
Workflow #1 ("Data Factory"): This process serves as the foundation of the entire system. It automatically accepts audio recordings, integrates with a transcription service, and then uses OpenAI to generate unique, customized summaries for each role. The final data is structured and stored simultaneously in PostgreSQL and vectorized for Supabase.
Workflow #2 ("Analytics Synthesizer"): Operating on a schedule, this workflow aggregates daily summaries from PostgreSQL, reuses AI to create a concentrated strategic report for the week for the owner, and automatically sends personalized briefings to key employees (Team Lead, BizDev) via Telegram.
Workflow #3 ("Interactive AI Assistant"): The pinnacle of the system — a Telegram bot that serves as a single access point to the knowledge base. I implemented:
Access control: The bot identifies the user and their role through the database, unlocking relevant features.
Admin commands: The ability to generate reports on demand by triggering Workflow #2.
RAG pipeline (Retrieval-Augmented Generation): A full-fledged question-answering mechanism. The bot converts the user's query into a vector, finds relevant information in Supabase, prepares context, and generates an accurate response using OpenAI.
Result:
Successfully developed and implemented an autonomous corporate "second brain" that operates 24/7. The client received a system that transforms conversations into structured assets, saving dozens of hours of work time and enabling data-driven decision-making rather than relying on memory.
The architecture is fully scalable: adding new roles, report types, or data sources does not require a system overhaul. The solution provides instant and secure access to information, ensuring that each employee sees only the data assigned to them.
#n8n #PostgreSQL #Supabase #pgvector #OpenAI #Telegram #TelegramBot #Automation #NoCode #API #APIIntegration #RAG #LLM #AIassistant #WorkflowAutomation #BusinessAutomation #Automation #ChatBot
Reviews and compliments on completed projects 11
15 January
226 USD
Tables, bots, automation
Quality
Professionalism
Cost
Contactability
Deadlines
Everything was done quickly and professionally.
|
27 October 2025
113 USD
Telegram bot + Google Sheets
Quality
Professionalism
Cost
Contactability
Deadlines
I recommend Mykhailo as a high-level specialist, with a deep immersion in details, understanding of processes, structured approach, and quick communication. Thank you for the work!
Yevhen Glazunov
| Personal | Response review
17 October 2025
27 USD
Fix errors in n8n scenarios and Supabase.
Quality
Professionalism
Cost
Contactability
Deadlines
The project is completed. I recommend for collaboration. Thank you.
15 October 2025
34 USD
Setting up a scenario in make.com - Integration of Google Sheets, OpenAI, and CRM
Quality
Professionalism
Cost
Contactability
Deadlines
Great understanding of Make.com, quickly grasped what needed to be implemented. The scenarios were set up neatly, logically, and without errors. All processes now work automatically, saving a lot of time. I especially liked that the freelancer was always in touch, explaining each step in detail. I recommend them as a professional, responsible, and pleasant specialist to communicate with!
Dmitry Spetsprom
| Personal | Response review
31 July 2025
226 USD
Integration of Notion and Airtable
Quality
Professionalism
Cost
Contactability
Deadlines
The task was completed. Thank you for your work
Nikita Butenko
| Personal | Response review
Quality
Professionalism
Cost
Contactability
Deadlines
By choosing Mykhailo, you choose quality and an individual approach. It was Mykhailo who approached the task decisively and in detail. Conducted a good meeting to identify all the pain points of the process, proposed better solutions than the initial concept. Helped effectively save money on services and subscriptions. And most importantly - automated processes that took 20 times more manual work. We outlined further cooperation to expand automation to other processes. The final guide on the work done, usage instructions, ongoing support - all of this remained a pleasant bonus! I recommend to everyone and anyone who wants their business to be efficient.
7 June 2025
16 USD
System audit based on Trello/Zapier/Make
Quality
Professionalism
Cost
Contactability
Deadlines
Mikhail handled the audit of our technical system perfectly. The approach was very structured: he quickly understood the essence, identified key problem areas, and proposed a clear and realistic plan to address them. Everything was presented clearly, with priorities and recommendations for the next steps. I recommend him for any tasks related to process optimization.
Filipp Lysak
| Personal | Response review
28 May 2025
16 USD
Make.com Setup for Data Storage and Basic Logic
Quality
Professionalism
Cost
Contactability
Deadlines
Thank you to Mikhail for the completed project. Saved a very significant amount of money. Highly recommend for collaboration.
Maksim Getmansky
| Personal | Response review
23 May 2025
113 USD
Development of a Telegram bot (sales funnel)
Quality
Professionalism
Cost
Contactability
Deadlines
It was very nice to work with Mikhail
He completed all the tasks clearly that were necessary and additionally provided consultations and tried to help me as much as possible
I will recommend him to all my acquaintances who have such tasks
23 May 2025
68 USD
Part 1.1: Setting up Manychat and Basic Data Collection
Quality
Professionalism
Cost
Contactability
Deadlines
Everything quickly and according to the technical specifications. Will recommend
Maksim Getmansky
| Personal | Response review
29 April 2025
16 USD
Consultation on Telegram bot
Quality
Professionalism
Cost
Contactability
Deadlines
Very cool specialist
Clearly explained everything without fluff