AI and Machine Learning
Retrieval, agents, and the plumbing underneath them.
- LLMs
- RAG
- LangChain
- AI Agents
- Prompt Engineering
- Vector Embeddings
- Vector Databases
- Semantic Search
- OpenAI API
- Model Inference
- n8n
I work at both ends of a product. Schema design, REST boundaries, authentication and query plans at one end. Retrieval pipelines, agent workflows and model inference at the other. Most of what I have released, I owned alone, from the first migration through to the production page at 2am.
Sole developer on AI SaaS products at NESL-IT, from system design through to release.
Retrieval over PostgreSQL holding millions of records. Plain language in, answers out.
SSR and SSG front ends measured on Core Web Vitals, in the field rather than on my own machine.
Omnichannel support desk. Email, live chat, Instagram, voice and Shopify threads land in one inbox, with response metrics, reply suggestions and role based access over the top.
Workspace for sales development teams. Prioritized call queues, sequences that mix email and calls, calendar handoff to account executives, and pipeline numbers that move as the day runs.
Reply platform running model inference pipelines. Personas set the tone, saved replies cover the repeats, and full proposals generate from a single brief.
Async workspace for engineering teams. Written standups roll into daily and weekly digests, repositories index into onboarding guides, and questions about the codebase come back cited to file and line.
Daily tools first, then the ones that come out when a problem calls for them.
Retrieval, agents, and the plumbing underneath them.
Rendering strategy first, then the interface.
Boundaries that hold when one service grows faster than the rest.
Schemas designed once, queried a million times.
Releases that repeat without a person watching.
Orders and inventory moving between systems on their own.
Email reaches me fastest. I reply inside a day.