Arbab C.

I put AI on top of enterprise data without breaking what already works — RAG, Claude/MCP agents, and the pipelines under them. Senior architect at a top-10 US bank (Federal Reserve regulatory data), 23 years of Oracle/SQL, three shipped AI products. Most of my clients come to me with one of three problems: • You want AI on your documents or your database, and the demo was wrong — made-up answers, missing citations, a retrieval layer that can't tell two similar records apart. I've built the retrieval layer three different ways and measured each; the honest finding is that most "AI" failures are data failures, so I start with the corpus and the schema, then the model. • The data has to move first — Excel to a real database, one system into another, three apps that disagree on the numbers — and it has to reconcile on day one. I build pipelines with the checks built in, because I spent years on systems where a wrong number is a regulator finding, not a bug. • Something in Oracle or SQL Server is slow, fragile, or wrong, and nobody can say why. I find the cause (usually the access path, not the hardware), fix it, and leave a note so it stays fixed. Latest: a bank's financial report from 30 seconds to 0.2. Background, briefly: accounting degree, CISA, two decades on mission-critical Oracle systems in banking, healthcare, and manufacturing, most recently designing regulatory-reporting data platforms (Basel, liquidity, Dodd-Frank). I read financial requirements as fluently as I read an execution plan — and I've shipped AI products of my own, so I know what production AI needs from the data underneath it. How I work: a small first milestone so you can judge the work before committing to more. Assessment first, then the build. Part-time, remote, Eastern US — I overlap comfortably with West Coast afternoons and I write things down so your team can run what I leave behind. If it's a chatbot demo, I'm not the right person. If it's AI that has to answer from your real records and survive an audit, send me the corpus and the current architecture and I'll tell you what I'd look at first. Stack: Python, FastAPI, Claude and OpenAI APIs, MCP, RAG (embeddings, pgvector, evals), Oracle, PL/SQL, SQL Server, PostgreSQL, ETL/ELT, data modeling.