GenAI Systems & Integrations
RAG search • Tool-using chatbots (MCP/function calling) • Light tuning • Guardrails & cost control
Proof of capability (live)
SmartDocs — RAG over PDFs: Next.js + LangChain + Jina embeddings + OpenRouter (GPT-3.5) with file/page citations.
What I deliver
- RAG pipelines: ingestion, chunking, embeddings, retrieval, grounded answers with citations.
- Chatbots with tools (MCP/function calling) for search, DB queries, internal APIs — demo available on request.
- Light tuning (LoRA/prompt frameworks) for tone & domain fit, when RAG alone isn’t enough.
- Safety & quality: guardrails, offline evals, usage budgets, and observability from day one.
Typical stack (options)
- Models via OpenRouter (OpenAI/Anthropic, etc.) or local (Ollama) where applicable.
- Vector stores: Pinecone / Weaviate / PgVector — chosen per scale & budget.
- Next.js or Node services with CI/CD, metrics, and cost governance.
Want reliable AI that ships?
I run short discovery → pilot sprints to land measurable outcomes (better search, faster support, content acceleration), with safety and costs under control.