AI application developer — RAG & agents
- £83-£137/hr
- Part time
- 9 years
- RAG pipelines
- LLM evals & guardrails
- Chatbot development
- LLM API integration
- Supabase
- Vector search & embeddings
- Prompt engineering
Projects 2
Documentation assistant
Part of a team of three
RAG over 4k support articles with a 120-case eval set. Answer accuracy 71% → 89% after chunking and reranking changes.
- llm-evals
- vector-search
- llm-integration
- supabase
Support bot with guardrails
Tech lead
Refusal and escalation rules so the bot declines policy questions instead of inventing answers.
- supabase
- vector-search
- chatbots
- llm-integration
About
Retrieval systems that can be evaluated rather than demoed. pgvector on Supabase, LLM APIs, retrieval tuning, and an eval harness so improvements are provable. I care about refusal behaviour as much as recall.