← Home

For consultancies, dev shops and product teams

AI agent development and LLM engineering, on contract

I take contract work with teams that need an engineer who has put AI agents and LLM pipelines into production and kept them running. Python and Go, on Google Cloud and AWS.

AI agents and multi-step pipelines

• Multi-stage pipelines with a guard at each stage, where a person signs off before anything is committed. Case study
• I run my own multi-agent workstation: several long-running Claude Code agents on Google Cloud, reachable from a browser or a phone.
• Health checks that catch an agent that looks alive but is stuck, such as one waiting on a prompt while its service reports active.
• I measure a tool design before choosing it. One comparison showed a small Go command-line tool returning 8.8 times less output per call than the equivalent MCP server.

Retrieval at scale

• Prepared and optimized a retrieval corpus of more than a million news articles on PostgreSQL, tuned for low-latency retrieval.
• Built a retrieval configuration aimed at one end client, a Global 100 law firm, around economic, political and government sources.
• Ran daily scrapers and ingestion pipelines across hundreds of sources.
• Translated hundreds of thousands of foreign-language articles with LLMs.

Document extraction and structured output

• A pipeline that identifies and reads about 20 tax form types from PDFs and images, with a JSON schema per form. Case study
• Transcript to a strict 41-field schema with schema-constrained decoding. Unsupported fields stay blank. Case study
• Natural-language search where the model writes a filter and Go code executes it, so no record is sent to the model to pick matches.

Running systems in production

• Traced 350 GB of database egress to a single query with pg_stat_statements and cut rows per call from 514.8 to 95.
• Took a hot query from 21.9 ms to 0.071 ms with a partial index.
• Vector search with pgvector and HNSW, embedding-based clustering, and an LLM judge on borderline merges. Case study
• Cloud Run services and jobs, Cloud Build, Cloud Scheduler, Secret Manager, tracked migrations and CI.

Stack

Python, Go, JavaScript and SQL. Google Cloud (Cloud Run, Vertex AI) and AWS (Lambda, EC2, API Gateway). PostgreSQL, pgvector, Docker, GitHub Actions.

How I work with a team

Hourly or a fixed price for a defined deliverable. US Eastern time. I can start now.