Available for work
I take on a small number of engagements at a time — usually where AI automation and systems reliability overlap, and where the problem is hard enough that a template won't do.
Services
Projects
How it works
-
Scope call
Thirty minutes on what you're actually trying to achieve. If I'm not the right person, you'll know by the end of it.
-
Written proposal
Fixed scope, fixed price, explicit deliverables. No open-ended hourly billing.
-
Build in the open
Weekly updates and a working environment you can look at from week one — not a black box at the end.
-
Handover
Documented, tested, and yours. I leave when the system runs without me.
- Problem
- The agent already runs, but routes requests to the wrong place, calls the wrong tools, loses state, loops, or gives unstable answers.
- Result
- Root-cause audit, reproducible tests, the fixes themselves, error handling, metrics, and a report on the risks that remain.
- Stack
- LangGraph, Python, OpenAI/Anthropic, pytest, OpenTelemetry, Langfuse.
- Problem
- The bot retrieves the wrong documents, hallucinates, cannot cite its sources, or shows information to the wrong user.
- Result
- Retrieval diagnostics, access-control verification, an evaluation dataset, an improved pipeline, and regression tests to hold it there.
- Stack
- PostgreSQL/pgvector or Qdrant, BM25, RRF, reranking, RAGAS, permission-aware retrieval.
- Problem
- Staff sort leads by hand, update the CRM, draft emails, and pass requests along to each other.
- Result
- Automation from the incoming event to the agreed action, with error handling, logging, and protection against duplicates.
- Stack
- n8n, Python, REST, OAuth, CRM APIs, PostgreSQL.
- Problem
- A client needs an agent wired into internal APIs, and that requires access limits, validation, audit, and safe execution.
- Result
- An MCP server, tool schemas, authorization, tests, and documentation.
- Stack
- Python/TypeScript, MCP SDK, OAuth, policy enforcement, structured logging.
- Problem
- A company wants to ship an AI feature but does not know which models, components, limits, and operational machinery it actually needs.
- Result
- Architecture review, threat model, a quality-evaluation plan, a cost estimate, and a rollout roadmap.
- Stack
- LLM APIs, LangGraph, RAG, Docker, PostgreSQL, OpenTelemetry.
Support without a human in the loop
Replaces simple rules-based chatbots with agents that do not just answer the question but carry out the action inside internal systems.
B2B sales
Replaces the first-contact SDR. The agent enriches the company record, scores the lead against a scoring model, and books the call itself.
Accounts payable automation
Automates the work of accounting and procurement: reading awkward PDFs, scans, and receipts, and posting the extracted fields into the ERP.
Resume screening and first interviews
Speeds up hiring — working through hundreds of applications, matching them against a complex role profile, and running a basic technical and HR screen.
SOC / AI security
Automates the work of security engineers: agents triage thousands of security alerts a day and filter out the false positives.
GraphRAG
Replaces plain keyword search with a research agent that finds the hidden connections across thousands of company documents and builds sourced reports from them.
Internal ops
Removes bureaucratic busywork inside IT, FinTech, and retail companies when a new employee joins or a resource is requested.