Six clearly defined practice areas. Every project is scoped up front — no surprises later.
Assistants with personality, context and memory. Embedded where your users already are.
I design assistants that understand your domain and hold useful conversations instead of generic replies. I work with OpenAI, Anthropic or open models depending on the use case, with guardrails, evals and a panel for prompt iteration. Deployment to web, WhatsApp, Slack or your own channels.
Semantic search over your docs, contracts, manuals or knowledge base.
I implement complete RAG pipelines: ingestion, document-aware chunking, embeddings, vector store, reranking and answers with citations. The goal is not to point a model at a PDF — it is to build a system that knows when to say "I don't know".
AI capabilities inside your product without rewriting it.
I add AI to applications already in production: summarization, classification, structured extraction, better search, generation. The integration is designed to be controllable, observable and reversible. I meet your stack where it lives — Node, Python, PHP, Go, whatever fits.
Agents that run multi-step tasks with tools and human supervision.
I build agents that combine LLMs, tools and memory to automate real work: ticket triage, document processing, research, internal ops. With traceability, capped costs and human checkpoints where they matter.
Before building anything, deciding what is worth building.
Working sessions to map use cases, evaluate technical feasibility, estimate operating costs and pick a model. Audits of existing AI projects: prompt quality, cost, latency, evals, security and maintainability.
Wire your internal systems into Claude, ChatGPT or other clients via Model Context Protocol.
I design and build MCP servers that expose your APIs, databases and internal tools to AI assistants in a secure and traceable way. Useful for teams that want their internal processes accessible from Claude or ChatGPT without rewriting anything.
Book a free 30-minute call, or send me a message.
