Services
Services
AI / RAG products
I build AI assistants that answer from a real body of documents and show where each answer comes from — end to end, from ingesting the sources to the app, the API and the server it runs on.
What's included
- Document ingestion: PDF parsing, chunking, metadata
- Hybrid retrieval (dense + sparse) with reranking over a vector store (Qdrant/ChromaDB)
- Answers that cite their sources
- Usage tracking and cost-per-question control
- Multilingual products (Romanian, Russian, English); Next.js front end, Python/FastAPI services
Related work
FAQ
What kind of AI systems do you build?
Assistants where accuracy and sources matter: legal research, internal knowledge bases, domain-specific Q&A. Legy.md, an assistant for Moldovan law, is my own product.
Which models do you use?
Mostly OpenAI's API, with retrieval and prompts tuned to the documents; I also work with Anthropic's API and choose by quality and cost for the use case.
Can it work in Romanian and Russian?
Yes. Legy answers in Romanian, Russian and English.