Ross F.

I connect AI agents (Claude, GPT, Gemini) to a company's own data, and I measure whether their answers are right. Mostly that means MCP servers (how agents reach your tools and data) and RAG pipelines (answers grounded in your documents), which I build and run end to end as the one senior backend engineer on the job. Recent result: took a client's AI document-extraction accuracy (F1) from 43% to 84%, measured against the registries' own published data. Top Rated Plus, 100% Job Success, $300K+ earned, 8,500+ hours. RECENT WORK • Carbon-markets data platform (MCP server). The client's users ask an AI agent about 26K carbon projects across 20 registries and standards in plain English. Runs on AWS behind their agent service; every tool is read-only and credential-gated, and model or prompt changes are A/B-tested against a 249-scenario eval suite. The data underneath comes from pipelines that have collected 250K+ documents, including from sites behind Cloudflare and Incapsula. • Same client (document extraction). A RAG service that pulls structured data out of messy registry PDFs. Accuracy (F1) on one extraction went from 43% to 84% against registry ground truth; a September 2026 eval on another cut false matches from 67 of 67 to 1 of 67 while still catching 98% of true matches. Every output links back to its source document. (FastAPI, Celery, Pinecone) • Healthtech startup (clinical research Q&A). Lead backend engineer on a knowledge-graph platform in production for multiple institutions, 15K research papers indexed. Answers carry numbered citations to the exact passages they drew on, and each customer's data is kept separate, with dedicated cross-tenant tests. (FastAPI, Neo4j, MongoDB) • Consumer savings platform, US (data backend). 30+ retail and banking sources normalised into one schema, 60K+ fresh records a day, with Prometheus/Grafana alerting into Slack. WHAT I'M HIRED FOR • Connecting AI agents to your systems: MCP servers, Claude and Claude Code integrations, custom agents, tool design, evals • RAG and document extraction with measured accuracy, source citations and model-cost control • The backend underneath: FastAPI/Flask, Celery/RabbitMQ/Redis, PostgreSQL/MongoDB/Neo4j, Docker/AWS • Getting data in: PDFs, spreadsheets, sources with no API, sites that block automation WHO I WORK WITH Data-heavy teams in carbon markets, healthcare and consumer finance who want one senior engineer to own the AI backend. About two-thirds of my hours come from three engagements of 18 months to 2.5 years. HOW I WORK • I own the system: architecture, build, deployment, monitoring, handover docs. • Tests on everything, evals wherever an LLM is involved, and alerting that pages me when something breaks. • If an LLM is the wrong tool for the job, I'll say so before you pay for it. • Fixed-price milestones for defined builds; hourly for ongoing work. From a client review: "[He] troubleshot problems and fixed them, completed the project on time, and saved me hours of manual work on a weekly basis... This is the guy you want." UK-based (GMT/BST). Send me two or three lines: what the agent or pipeline should do, what data it needs, and your current stack. I'll tell you quickly whether I'm a fit and how I'd approach it. Or book a 30-minute consultation from my profile.