Creation of a hybrid personalization mechanism for a web store based on edge technologies

About the Project We are looking for an experienced AI Automation Engineer to design and build a secure, self-hosted AI platform that combines a local Large Language Model (LLM), Retrieval-Augmented Generation (RAG), and multiple AI agents to automate business workflows. This is a hands-on engineering role for someone who has experience building production AI systems—not simply integrating ChatGPT APIs. The goal is to create a private AI ecosystem capable of securely indexing company knowledge, answering questions using cited sources, processing meeting transcripts, and automating internal business processes. Responsibilities You will be responsible for: Designing and deploying a locally hosted LLM on a VPS or dedicated server Building a secure RAG pipeline using frameworks such as LlamaIndex or similar Creating document ingestion pipelines supporting PDF (including OCR), DOCX, TXT, XLSX and meeting transcripts Implementing document indexing, metadata management, deduplication, and versioning Developing AI agents for: Meeting transcription processing Automatic meeting summaries Action item extraction Client knowledge retrieval Building APIs or a simple web interface for querying the knowledge base Ensuring strict client data isolation and permission controls Implementing source-cited responses to minimize hallucinations Optimizing system performance, scalability, and reliability Writing documentation and deployment guides Performing testing and security validation Required Skills Strong Python development experience Experience with LLM frameworks RAG architecture experience LlamaIndex, LangChain, or equivalent Vector databases (Qdrant, Chroma, Pinecone, Weaviate, FAISS, etc.) Local/open-source LLM deployment (Llama, Mistral, Gemma, DeepSeek, etc.) API development (FastAPI preferred) Docker Linux server administration VPS deployment Git Authentication and access control Experience with OCR pipelines Experience working with structured and unstructured documents Fluent English What We’re Looking For The ideal candidate: Has built production AI systems from the ground up Understands RAG best practices Can work independently Thinks like a software architect—not only a developer Writes clean, maintainable code Communicates clearly Can recommend the best technologies instead of simply following instructions Project Type Freelance / Contract Remote Milestone-based Long-term opportunity for future AI automation projects Please Include With Your Application Portfolio of similar AI/RAG projects Examples of local LLM or AI agent implementations Estimated timeline Estimated project cost Hourly or fixed-rate pricing

I'm looking for a developer to create a Telegram bot for our clothing store. The essence: the manager sends the bot a regular photo of the product (on a hanger, on a table) — the bot generates a ready gallery of professional shots for the product card on the website/marketplace using AI. What the bot should be able to do Receive 1–4 photos of the product in the Telegram chat. Automatically recognize the product (category, cut, male/female) — without manual description. Generate a gallery of ~6 shots: product on a model: studio front, back, dynamic pose; product shots: flat lay, close-up detail, ghost mannequin; lifestyle shot. Consistent model — the same person in all shots of the gallery and between galleries (brand face). Several shooting styles to choose from (for example: streetwear, casual, sport, old money). Output format: JPG 4:5, files with numbering (-01, -02…) — ready for upload. The ability to regenerate a single shot or change the style without regenerating the entire gallery. Access only for our team (allowlist), daily limits per user. Accounting for AI API costs: show the price before generation, keep statistics. Technical expectations Image generation: Gemini (Nano Banana) / DALL·E / Flux or similar — please suggest an option with justification of quality and cost per shot. The bot must operate 24/7 on VPS (auto-restart, turnkey deployment). Secrets via .env, code delivered to us (repository). What we will provide Product photos for testing, wishes regarding brand styling. What I expect in the response Cost and timeline estimate. Approximate cost of one gallery. From and To

Auto-notification system for inventory replenishment analytics. On 8n8. Small automation for e-commerce. Once a day, the system calculates the sales velocity for each product and sends a message in Telegram indicating when to reorder (with the recommended quantity) and how many days the remaining stock will last. What we do: Data retrieval: KeyCRM API (order history) + stock/orders from Rozetka (API or export) Storage in SQLite on my VPS (Contabo, Linux) Calculation of average sales, days of stock, reorder point ABC analysis: prioritization of which product to replenish first (class A — main by revenue — more important) Daily Telegram bot + cron job All thresholds (lead time, safety stock, target stock) — in the config

A system is needed for the automatic processing of large volumes of information using AI. I am currently using my own prompt in Claude. It works well with small texts, but with larger volumes, the neural network may skip parts, stop before the end, or report completion even though the material has not been fully processed. A system is needed that can accept a book of 200+ pages, several books, PDFs, DOCX, TXT, subtitles, transcripts of audio and video, or another large array of information and automatically process it from start to finish. One of the main formats for the result for each meaningful sentence: English version. •••••••••••••••••••••••••••• Russian version. •••••••••••••••••••••••••••• English version. •••••••••••••••••••••••••••• English version. •••••••••••••••••••••••••••• Russian version. •••••••••••••••••••••••••••• English version. Each meaningful line is output four times in English and two times in Russian in the specified order. The translation must accurately convey the meaning. Short sentences can be combined, and long ones can be divided into complete meaningful lines. The format must be customizable: the number of English repetitions and translations, their order, languages, separator, line length, main prompt, and saving different templates. For example, instead of English–Russian–English–English–Russian–English, any other sequence can be chosen. If the document cannot be processed in one request, the program should automatically split it into internal blocks, send them to Claude, ChatGPT, Gemini, or another model, check the result, repeat problematic parts, and combine everything into one file. The user should not have to manually copy text for four pages. It is necessary to check that no sentence, paragraph, or meaningful fragment is missed, that there are no duplicates between blocks, and that the structure corresponds to the template. The check should not rely solely on the neural network's assertion. Audio processing The system should also process large archives of audio, such as a Telegram channel with 300 broadcasts of about one hour each. It is preferable to automatically download audio or accept the entire archive, recognize speech, and process everything without manual involvement. A structured summary should be created for each broadcast: topic, main thoughts, important facts, examples, recommendations, and conclusions. Greetings, advertisements, conversational filler, and meaningless repetitions are removed, but useful information is retained. After processing, not only separate notes are needed, but also one comprehensive readable document where the information is organized by topics. If a topic was discussed in different broadcasts, the materials are gathered into one section, duplicates are removed, and links to the original recordings are preserved. The results should include: — a summary for each broadcast; — a general thematic document; — search by words and topics; — connection of conclusions with the original audio; — saving progress and continuing after errors. I can already perform most of these actions myself, but only on a small scale and manually. Therefore, I am also open to considering a more efficient way to download and convert audio to text if the contractor offers a solution better than what I currently use. What to include in the response Please write: — how the system will be implemented and what product I will receive; — which AI and speech recognition models will be used; — how completeness is checked and how omissions and repetitions are excluded; — whether prompts, the number of translations, repetitions, and their order can be changed; — whether it is possible to download materials from Telegram; — the cost of development, API, timelines, and the price for further improvements. The main goal is a universal system that processes large books and hundreds of hours of audio without manual involvement, does not lose information, and delivers a finished result strictly according to the chosen template.

We specialize in obtaining grants for existing companies that plan investments. We came up with the idea of creating/connecting available AI models to search for companies that are planning or are already in the process of investing and to offer them grants for this purpose (we mainly deal with grants for the purchase of machinery and equipment). The AI would search available websites and registries to find investors. Then, it would also handle the automation of mailing to these companies and would filter out companies that: - fit into the lead categories (I think, 5-6 parameters); - replied to the email and are interested in the grant; Of course, additional suggestions are welcome; I am also considering Voice Calling and SMS sending. I would like to emphasize that I am not only interested in finding LARGE or very LARGE companies, but also smaller ones, for example, employing 8-50 people. And not just large cities, but also smaller towns. Sample target for testing: SUBCARPATHIAN For micro, small, and medium-sized enterprises (SMEs) and large enterprises from the Subcarpathian Voivodeship that belong to the security and defense industry. An enterprise from the security and defense industry is understood as an enterprise that: develops or manufactures products, technologies, or systems intended for military applications, related to national security, critical infrastructure protection, or citizen safety or provides services supporting security and defense systems (e.g., in the field of cybersecurity) or conducts research and development in the area of military technologies and security systems and which achieves a significant share of revenue from the sale of products, technologies, systems, or services in the area of security and defense, i.e., above 5% of total net revenue (cumulatively over the last 3 closed accounting periods). Support available for companies that conduct business in an organized and continuous manner in the Subcarpathian Voivodeship for a period of no less than 24 months counting back from the date of the call for applications. We have proposals for websites to search. Ultimately, the program would prepare a daily summary of such investors for us. Can someone help us with this? I read that it is possible, but I don't have time to deal with it myself :)