www.ai.rs
AI & ML interests
- Natural Language Processing - Text Generation - Conversational AI - Fine-tuning - Retrieval-Augmented Generation - Reinforcement Learning from Human Feedback - Quantization - E-Commerce - Recommendation Systems - Edge AI
title: README
emoji: ๐ค
colorFrom: blue
colorTo: cyan
sdk: static
pinned: false
๐ค ai.rs โ Custom AI for E-Commerce
We build AI sales assistants for e-commerce businesses. Our models know your products, speak your brand, and work 24/7.
๐ง What We Build
- ๐ง Fine-tuned LLMs โ Custom-trained Qwen, Llama, and Mistral models on your product catalog
- ๐ RAG Pipelines โ Real-time retrieval from your database for accurate pricing, availability, and specs
- ๐ฌ AI Sales Assistants โ Embedded chat that recommends products, answers questions, and drives conversions
- ๐ฅ Dedicated Deployment โ Your own server, your own data, no shared infrastructure
โ Our Approach
- Data preparation โ We structure your product catalog, FAQs, and sales conversations into training data (~26,000+ examples per deployment)
- Fine-tuning โ We train on open-source base models (Qwen 2.5, Llama 3, Mistral) using your domain-specific data
- Safety training โ DPO alignment to keep responses on-brand and prevent hallucination
- RAG integration โ Vector search + live database queries for real-time pricing and inventory
- Deployment โ Dedicated GPU server running vLLM, with monitoring and updates
๐ Tech Stack
| Component | Technology |
|---|---|
| Base models | Qwen 2.5 (8B), Llama 3 (8B), Mistral (7B) |
| Fine-tuning | LoRA / QLoRA, custom training pipelines |
| Serving | vLLM on dedicated GPU servers |
| RAG | Vector embeddings + live SQL/API queries |
| Infrastructure | Rust + Python |
๐ Based in Belgrade, Serbia
Serving e-commerce businesses across Europe and beyond.
๐ ai.rs ยท ๐ง github@ai.rs ยท ๐ GitHub