Create an AI agent for inventory management

Budget: 6000 UAH Deadline: 14 days

Good day! Your task is what I do every day: I keep agents on Claude engaged with operational tasks, so I approach this from a practical work perspective rather than a presentation.

Here’s how I see the first stage:

1) Data exchange. We take QVD from QlikView every hour to your VPS and distribute it into a normalized database (Postgres). The agent works with this slice, not with 30 GB of raw data, so responses are quick and predictable in cost.
2) Agent. A separate profile of conditions (processing times, logistics, batches) for each supplier: a prompt plus strict rules. The agent prepares the request, and a person confirms it in Telegram. Purchases do not proceed without confirmation.
3) Movement between warehouses and tasks by groups of goods (cutting, selection by lengths) work with the same mechanism.
4) Telegram: several employees, roles, the task goes to a specific person.
5) Bitrix24: task creation via REST with a callback link.
6) Dashboard: how many analyses, how many tasks set and closed, list of goods in procurement.

I propose the first stage as a pilot with 2-3 suppliers: exchange, database, agent, Telegram, Bitrix, dashboard. The accuracy is immediately visible with real prompts, and only after that we can honestly calculate the rollout for all 40+.

Question about the data: do we read QVD directly or can QlikView simultaneously export the same slices in CSV/QVX? This affects the exchange layer.

About me: 5 completed projects on Freelancehunt, 5 reviews of 10/10, no negative feedback. Plus a public history, 8 merged PRs into other open-source projects, mostly in a Go security tool with 173 stars; each passed maintainer review.

Pilot 6000 UAH, 14 days. I set the starting price: I am interested in a long project, so I am going in favorably at the first stage. I am ready to show live bots right in the chat so you can see the level before the solution.

Petro Pankov, BotCraft Group Expand

Budget: 20000 UAH Deadline: 15 days

Hello, Vitaly! We have already discussed your SQL database in private — here’s how I see the implementation now that we have the complete technical specification:
Stack: Python + FastAPI, Claude API with tool-calling (the agent calls specific tools — create an order, set a task in Bitrix24, rather than just generating text), PostgreSQL for the state of the agent/tasks, aiogram for Telegram with employee segregation.
As we discussed: I will connect to the SQL database read-only directly (without parsing QVD through the QlikView layer, as most suggest), pulling only the necessary tables (stocks/orders/suppliers). This removes the unnecessary QVD/QlikView layer entirely, and the agent works with current data directly.
Stages:
Stage 1 (MVP): analysis of stocks/orders through Claude, formation of orders for 2-3 suppliers for the pilot, Telegram bot with confirmation from the employee (not auto-sending directly to the supplier — safer at the start). — 20,000 UAH, 14 days.
Stage 2: other suppliers, transfers between warehouses, tasks for cutting, Bitrix24 integration, dashboard — we will discuss after the pilot.

Budget: 2000 UAH Deadline: 2 days

Hello!
I am Oleg, project manager and solutions architect at Business Atlas (partners of n8n). The main focus of your brief is the secure processing of QVD files from QlikView, taking into account the unique conditions of 40+ suppliers and integration with Telegram and Bitrix24.
1. Our vision of architecture and approach:
Orchestration (Core): Self-hosted n8n in Docker on your VPS. Automatically reads hourly QVD files (250 MB), imports data into an intermediate database (PostgreSQL/Supabase) via a Python script for AI processing.
"Non-destructive first" principle: AI creates proposals in draft status ("AI Draft"). The manager approves the order in the Telegram chat (with "Accept/Edit" buttons), and only after that n8n assigns the task in Bitrix24 and sends the brief to employees.
Knowledge base (Claude + RAG): We upload 40+ supplier conditions into a vector storage (Qdrant). Claude analyzes the context of a specific supplier and formulates orders according to their rules.
Dashboard: a panel in Looker Studio or Google Sheets (metrics of analyses, tasks, CRM, and lists of products for procurement).
2. Relevant experience:
AI infrastructure (1700+ SKUs): self-hosted system (n8n + Ollama + LlamaIndex) with moderation via Telegram based on the "Non-destructive first" principle.
Data conversion and unification: automatic parsing and normalization of heterogeneous supplier files through an AI pipeline and Qdrant.
Direct connection to the database: AI agent works directly with the local ERP system database (Firebird) without a ready REST API.
TRUST Telegram bot for logistics: a complex bot (Python, aiogram) with multi-step scenarios and reporting.
3. Stages, timelines, and costs:
Phase 1: Paid Discovery (3–5 days) — $230 (test parsing of a QVD file, prompts for 1-2 suppliers, brief. The amount is credited towards development).
Phase 2: MVP of procurement and transfers (3–4 weeks) — $1,500 – $2,500 (n8n, reading QVD, database, agent logic, Telegram bot).
Phase 3: Integration with Bitrix24 and task splitting (7–10 days) — from $800 (Bitrix24 API, specific rules, dashboard).
Is it possible to send a sample of one QVD file for initial analysis of its structure?
Please write in private to receive a more personalized offer. Expand

Budget: 15000 UAH Deadline: 14 days

Hello! The project looks very large-scale and technological. I specialize in backend development in Python, integration of AI agents, and automation of logistics processes. I have experience in designing architecture for AI assistants, as well as creating monitoring systems for supply chains with inventory record synchronization, so I am very familiar with the logic of warehouse movements and procurement. Regarding technical implementation: QVD is a specific binary format for Qlik, which is difficult for external scripts to work with. The best and fastest solution would be to set up an automatic export of these 250 MB in CSV, Parquet, or lightweight SQLite format in your QlikView. I will build the agent in Python (using the LangGraph or CrewAI frameworks) with a connection to the Claude 3.5 Sonnet API. Since the data volume is large, the AI will not read the entire file directly (this is impossible due to token limits), but will work as an analyst—generating SQL or Python queries to the data and making decisions based on the results. The dashboard can be quickly deployed on the Streamlit framework, and interaction with Telegram and Bitrix24 via REST API. I am ready to review the structure of your data and discuss the agent's working scenarios!

Budget: 20000 UAH Deadline: 14 days

Hello!

I specialize in automating business processes and implementing AI agents. I work with n8n, Python, CRM systems, API integrations, and LLM (OpenAI and others). I help reduce routine tasks, optimize processes, and increase efficiency.

I am ready to quickly and efficiently complete your task. Let's discuss the details!

🎨 https://gurtop.com/

Budget: 1000 UAH Deadline: 1 day

Good day, Vitaliy.

Since you have over 40 suppliers with different conditions, the key is not the agent itself, but the data structure that allows it to make decisions without the risk of errors in real orders. QVD files require preliminary conversion into a format that Claude works with reliably — and this transitional layer will determine the accuracy of the entire process. Instead of manually writing prompts for each contractor separately, it makes sense to build a rule system where supplier conditions are structured as data rather than as text in instructions. This significantly reduces the risk of hallucinations when forming orders and tasks in Bitrix24, and the dashboard will provide transparency for each agent's decision — how many analyses have been completed, how many tasks have been created, and their status.

Stages:

Each stage is documented, and the code is passed on to you.
You can stop after any stage.

≈12 days: 11,000–15,000 UAH
Minimally viable agent for automating one procurement direction.
- Automated data import from QVD files into an accessible format
- AI agent for working with a limited group of suppliers (up to 5 people)
- Integration with Telegram for receiving needs notifications
- Basic functionality for task creation in Bitrix24

≈30 days: 25,000–35,000 UAH
Full-fledged procurement and movement management system with quality control.
- Full cycle of QVD data processing with hourly updates
- Logic setup for all major suppliers (up to 40+) with iterative testing
- Human action confirmation system before sending to CRM/Telegram
- Interactive dashboard with metrics of analyses and completed tasks
- Full functionality for managing movements between warehouses

≈45 days: 45,000–60,000 UAH
Extended complex with deep analytics and scaling support.
- Advanced data validation system for complete elimination of AI hallucinations
- Specialized module for calculating product splitting into parts
- Priority support and fine-tuning of prompts for 2 months
- Architectural preparation for creating a sales analysis agent

Required experience:
- https://valpere.github.io/portfolio/
- https://valpere.github.io/projects/
- https://valpere.github.io/mvb/ Expand

Budget: 27000 UAH Deadline: 30 days

Good afternoon!

For such a task, it is rational to divide the solution into several layers: loading and normalizing QVD, a history storage of decisions, a separate rules layer for suppliers, and an AI layer for analysis, generating task drafts, and explaining the results. Critical purchasing decisions should not be handed directly to the model: the agent should rely on formalized supplier conditions, stock levels, minimum batches, deadlines, and warehouse priorities, while LLM should be used for processing non-standard cases and preparing understandable actions for employees.

From a technical perspective, there are two main risks here. The first is QVD and hourly updates: a stable ETL process with incremental processing and caching of aggregates is needed; otherwise, the system will be unstable with large volumes of data. The second is heterogeneous supplier conditions: if they are not placed in a separate rules layer, the quality of recommendations will be variable. For Telegram and Bitrix24, it is essential to immediately incorporate roles, an action log, and confirmation of disputed operations by a person.

Claude seems to be a reasonable option for the language part, but architecturally, it is more important not to choose the model, but the scheme around it: a task queue, version control of prompts, storage of decisions, a test circuit on real scenarios, and a dashboard for analyses, tasks, and execution statuses. This way, it will be easier to scale the system to the next stage—sales analysis—without reworking the foundation.

For an accurate assessment, it is critical for us to see an example of the QVD structure and understand which actions the agent performs automatically and which it only suggests for confirmation. It is also important to establish the boundaries of the first stage: we start with purchases and internal transfers or immediately incorporate a dashboard and integration with Bitrix24.

Budget: 27000 UAH Deadline: 16 days

Hello! The project is extremely interesting and has a clear engineering task.

The main technical nuance here is 250 MB QVD files every hour. This is a binary format of QlikView, which is too large for direct transfer to the Claude prompt (it's expensive and clogs the context window).

I propose a fully autonomous and reliable architecture on the Laravel + Filament PHP stack, which will work with your files "as is" — you do NOT need to involve your QlikView administrators to reconfigure the export or write additional scripts.

How this will be implemented:

1. Autonomous QVD pipeline (on VPS):
- A fast CLI converter (compiled binary) is set up on the server, which unpacks your QVD files in seconds every hour.
- Laravel Scheduler streams (without overloading RAM) loads and updates stock balances and product movement in the local database.

2. Core AI Agent (Laravel + Claude API):
- Mathematical calculation: analysis of critical stock balances, reorder points, and need for movement between warehouses is performed deterministically at the database level.
- Claude API module: processing unique conditions for each of the 40+ suppliers (from a flexible base of prompts/rules), generating texts for orders and tasks for processing/splitting.

3. Integrations:
- Telegram bot: for convenient multi-user work of employees (notifications, order confirmations, operational commands).
- Bitrix24 REST API: automatic creation and assignment of tasks to the relevant managers in the CRM.

4. Monitoring dashboard (Filament PHP):
- Convenient web interface for control: number of analyses performed, list of assigned tasks, execution status, CPM, and overall list of products in procurement.

Estimated development time for MVP: 14–18 days
Cost: 50,000 UAH

I have extensive experience in building complex backend systems on Laravel, integrating LLM (Claude API), developing Telegram bots, and REST APIs. I am ready to discuss the details in private messages! Expand

Budget: 14700 UAH Deadline: 5 days

Hello! I am Serhiy, I have 8 years of experience in IT across various technologies: web/mobile, backend, API, integrations, automation, and data handling.
I see the main task as follows: Create an AI agent for inventory accounting. A manufacturing company with several warehouses in Ukraine. More than 40 suppliers, both within the country and in China....

Skills for the task:
- bots, parsing, and data handling
- reporting, analytics, and business automation
- solution architecture
- development, testing, and release preparation
- post-launch support

Portfolio: https://software.campstudio.net/#showcases Expand

Budget: 27000 UAH Deadline: 15 days

Good day.
Our team has many years of experience in developing ERP, CRM, CMS, and specialized software for businesses. We create effective digital solutions that help automate processes, increase productivity, and scale companies.

We work with modern technologies — from bots and scripts to AI agents and analytical systems. We develop websites of varying complexity. In our portfolio, we have implemented ERP solutions for the hospitality industry, as well as for companies engaged in the import and sale of goods, and our own product XFitness — an ERP system specifically designed for fitness clubs.

We are ready to implement your project and offer the best solution tailored to your needs.
Our portfolio: Freelancehunt

We specialize in the following areas:
- Development of ERP Systems
- Development of CRM Systems
- Development of Websites of any complexity
- Development of CMS Systems
- Support for Websites
- Development of OpenCart
- Support for OpenCart
- Modification of OpenCart
- Enhancement of OpenCart
- Development of WordPress
- Support for WordPress
- Modification of WordPress
- Enhancement of WordPress
- Development of ECommerce
- Support for ECommerce
- Modification of ECommerce
- Enhancement of ECommerce
- Development of Web Applications
- Support for 1C Servers
- Support for Web Servers
- Development of mobile applications
- Data parsing
- Development of bots
- Development of AI agents

and on the following technologies:
- Python
- PHP
- Laravel
- Symfony
- Yii2
- JS
- NodeJS
- jQuery
- TypeScript
- MySQL
- HTML
- CSS
- Vue
- Nuxt.js
- React
- React Native
- C++ Expand

Budget: 700 UAH Deadline: 4 days

Hello, I am in the TOP 10 freelancers on the platform, I have completed similar projects before, I will do it quickly, efficiently, and at a low cost.

Budget: 27000 UAH Deadline: 18 days

Good day.
Before proposing an implementation, I would like to familiarize myself with the structure of QVD files and examples of rules for at least a few suppliers. This will determine how best to build the agent's logic so that it works not only based on prompts but also consistently delivers predictable results.

I see that at the first stage, it is necessary to implement a complete cycle: obtaining data from QVD, analyzing inventory, generating purchasing and warehouse transfer recommendations, integrating with Telegram and Bitrix24, as well as a simple dashboard for monitoring the agent's performance. It is important to lay out the architecture in such a way that in the future, it will be easy to add sales analysis and other automation scenarios.

I work with Python, AI integrations, Telegram Bot API, and REST API. For such a project, I would use FastAPI, Claude/OpenAI (depending on the chosen model), a separate service for processing QVD, and Docker for deployment on VPS.

If possible, please send an example of one QVD file and describe the rules for at least one supplier - after that, I will be able to propose an optimal architecture and immediately start working.

Budget: 27000 UAH Deadline: 21 days

Vitaliy, good afternoon.

I am ready to take on the creation of an AI agent for inventory management and procurement.

I suggest starting with the first working phase: gather requirements, understand the current logic of procurement and movements, set up a pilot version of the agent with several suppliers, connect employees to work via Telegram, add task creation in Bitrix24, and prepare a simple dashboard for monitoring results.

It is important to make the system not just a "chatbot," but an assistant that provides clear recommendations, records tasks, and helps employees make faster decisions regarding procurement and inventory.

The cost and timeline I indicate are for the MVP/first phase. After reviewing the data, examples of suppliers, and real scenarios, it will be possible to accurately assess the full implementation across all suppliers.
Budget: 1000 Eur
Timeline: 21 days Expand

Budget: 15000 UAH Deadline: 5 days

Hello Vitaly.
My name is Sergey.
I would be happy to work on your project.
We will discuss the price.
Please write to me in private messages, I have a few additional questions.
The price and deadlines are currently approximate.

Best regards,
Sergey

Budget: 27000 UAH Deadline: 20 days

Good day! I am ready to develop an AI agent in Python for automating purchases and movements between warehouses. I will implement data processing from QVD, rules for working with each supplier, integrations with Telegram and Bitrix24 API, task management, and a simple web dashboard. In the first stage, I will analyze the structure of QVD and prepare the architecture of the solution on Claude API with deployment on your VPS. I have experience in developing Python services, AI agents, Telegram bots, API integrations, and business process automation systems.

Budget: 27000 UAH Deadline: 21 days

Hello!

I am interested in the project.

I have over 10 years of experience in developing complex web systems, automating business processes, integrating APIs, and working with AI.

After reviewing the description, I see that the optimal solution would not be "training" a model on QVD files, but rather building an AI agent that will work with current data, supplier rules, and business logic. This approach is significantly easier to maintain and scale.

I suggest starting with the design of the architecture and MVP, after which we can implement the functionality in stages.

After familiarizing myself with the current infrastructure and QVD files, I will prepare a detailed implementation plan and an accurate estimate. Expand

Budget: 27000 UAH Deadline: 20 days

From analyzing inventory to sending orders to suppliers and setting tasks in Bitrix — without manual intervention. I built LLM pipelines based on the Claude API of this type: the agent reads data, makes decisions based on prompts, and calls the necessary tools — Bitrix API, Telegram, order form. The tool-calling pattern through Claude is perfectly suited for 40+ suppliers with individual conditions — each supplier is set up with a separate set of prompts and instructions.

Regarding the stack: FastAPI on your VPS, Docker, QVD parsing through a Python library, PostgreSQL for agent state and tasks. A Telegram bot on aiogram — multi-user management with role differentiation. A lightweight dashboard (analytics, tasks, statuses, list of goods in procurement) on the same backend.

The cost is preliminary — we will finalize the figure after a short call to clarify the logic for each supplier, movement rules, and the structure of QVD files.

Write to me — we will discuss the architecture and the next step.

Budget: 15000 UAH Deadline: 8 days

Good day. I have already completed projects with AI integration. You can view the work in my portfolio. There are reviews for them in my profile. I am ready to implement, but we need to discuss details about suppliers and stores. Write to me, and we will agree.

Budget: 1500 UAH Deadline: 3 days

Hello! Am I correct in understanding that the key issue right now is adapting the agent to the individual conditions of each of the 40+ suppliers, and you would like to initially focus on one or two for a pilot launch?

I will detail the exact timelines and budget in a personal message.

Regarding implementation, I propose the following plan:
1. I will conduct an audit of the QVD file structure and determine the optimal data format for training the agent (Claude Cowork).
2. I will set up data exchange between QlikView and the agent via API, with minimal interference in your current infrastructure.
3. I will develop a system of prompts for each supplier separately, followed by testing in real procurement and movement scenarios.

Thank you for considering my proposal. I look forward to the opportunity to collaborate with you!

Budget: 20000 UAH Deadline: 14 days

Vitaliy, I see the task in automating procurement based on the analysis of QVD files that your QlikView server is currently collecting. The main challenge here lies in creating a layer that will effectively translate structured data from your reports into context for the AI agent, so that it takes into account the logistics of each of the 40 suppliers. My plan is to set up a data pipeline from QVD, prepare a vector knowledge base for Claude, and implement a prompt validation system so that the agent correctly generates orders without hallucinations. How critical is it for the agent to work with unstructured data from suppliers in the form of PDF price lists or email correspondence, or shall we limit ourselves to data from QlikView only?