Projects and articles related to AI/ML.
Apr 22, 2026
Platforms like LibreChat give you a production-ready chat UI, multi-model routing, and built-in Langfuse integration. Custom agent frameworks give you reactive UI, transactional tool execution, and full rendering control. Here is how I decide between them across four different products.
Apr 20, 2026
I built an AI evaluation framework from scratch to understand every piece — ground truth datasets, precision/recall harnesses, LLM-as-judge layers, ClickHouse dashboards. Then I migrated it to Langfuse. Here is why, and what I gained.
Apr 19, 2026
Part 1 of my portfolio AI series: Convex agent component design, RAG fundamentals, OpenAI vs Google provider testing, streaming UX, and how the frontend chat bubble and /admin/ai tooling live inside one reusable AI plugin.
Apr 18, 2026
Part 2 of my portfolio AI series: concrete eval results, ClickHouse monitoring signals, optimization decisions, recursion guardrails, and the measured before/after impact.
Apr 17, 2026
A practical guide to embedding dimensions using our Portal support retrieval pipeline and shared support/AI components: how dimension choice changes quality, latency, storage, and migration strategy.
Apr 16, 2026
Most developers use AI to generate throwaway code for every new project. I use it to build CLIs, agent skills, test harnesses, and plugin scaffolds — repeatable infrastructure that compounds. Here is why spinning up a fresh agent session per task, with pre-built tools and context, beats ad-hoc prompting every time.
Apr 15, 2026
Our first AI feature hallucinated a compliance finding that did not exist. A client acted on it. Here is how we built input sanitization, output filtering, confidence scoring, and human-in-the-loop escalation to make AI output trustworthy.
Apr 14, 2026
Our AI scanner said a page violated WCAG 2.4.7 — but it cited the wrong success criterion. RAG fixed the hallucination problem by grounding every AI claim in the actual specification, with retrieval metrics that prove the system works.
Apr 13, 2026
AI scanners find accessibility violations that deterministic tools miss — but how do you know the AI is not hallucinating? We built an evaluation harness that measures precision and recall against axe-core ground truth, gates every prompt change with regression tests, and tracks quality drift in ClickHouse.
Mar 31, 2026
Our first AI feature hallucinated a refund policy that did not exist. A customer followed it. Here is what we learned about putting language models in front of real people.
