Daniel Odiachi — Senior Full Stack & AI EngineerSenior Full Stack & AI Engineer with 8+ years shipping production systems. Building LLM products — RAG pipelines, agentic tool-use workflows, and a custom MCP server — on top of deep fintech infrastructure: cross-border payments, virtual card issuance, KYC/AML, and end-to-end encryption. Available for remote or on-site roles worldwide.https://danielodiachi.work/What it actually takes to ground a RAG support assistanthttps://danielodiachi.work/blog/grounding-a-rag-assistant/https://danielodiachi.work/blog/grounding-a-rag-assistant/Semantic search alone will not ground an LLM support assistant in a regulated product. Hybrid retrieval, exact-match recall, and a first-class "I don't know" are the parts that do the work.Mon, 20 Jul 2026 00:00:00 GMTRAGLLMRetrievalFintechHow to Evaluate RAG Without a Labelled Datasethttps://danielodiachi.work/blog/evaluating-rag-without-labels/https://danielodiachi.work/blog/evaluating-rag-without-labels/You will not get a golden set before you ship. Four evaluation techniques that need no labels — inverted queries, negative controls, retrieval-only metrics, and pairwise diffs — in cost order.Tue, 14 Jul 2026 00:00:00 GMTRAGEvaluationLLMRetrievalHuman-in-the-Loop Is an Architecture Decisionhttps://danielodiachi.work/blog/human-in-the-loop-agent-architecture/https://danielodiachi.work/blog/human-in-the-loop-agent-architecture/Approval prompts bolted onto an agent produce rubber-stamping, not oversight. A four-question framework for deciding which actions pause, and where the line between reads and writes really sits.Tue, 07 Jul 2026 00:00:00 GMTAgentsLLMArchitectureComplianceMCP vs. Bespoke Tool Integrations: When to Switchhttps://danielodiachi.work/blog/mcp-vs-bespoke-tool-integrations/https://danielodiachi.work/blog/mcp-vs-bespoke-tool-integrations/Hand-rolled tool definitions are correct for one assistant and wrong for three. A direct comparison of MCP against per-agent integrations, and the point where the tradeoff flips.Tue, 30 Jun 2026 00:00:00 GMTMCPAgentsLLMArchitectureStructured LLM Output You Can Actually Trusthttps://danielodiachi.work/blog/structured-llm-output-validation/https://danielodiachi.work/blog/structured-llm-output-validation/JSON mode and typed schemas stop parse errors, not wrong answers. The validation that matters happens after the model, and it's boring deterministic code you have to write yourself.Mon, 22 Jun 2026 00:00:00 GMTLLMValidationTypeScriptData ExtractionIdempotency for LLM Agents: Lessons from Paymentshttps://danielodiachi.work/blog/idempotency-for-llm-agents/https://danielodiachi.work/blog/idempotency-for-llm-agents/Payments engineering solved unreliable retries a decade ago with idempotency keys and a ledger. LLM agents have the same failure mode and almost none of the discipline. Here's the transfer.Fri, 12 Jun 2026 00:00:00 GMTLLMAgentsPaymentsReliabilityArchitecture