Beyond the demo: what production AI actually needs.
Notebook demos hide the failure modes production systems live and die by. A working checklist for turning a model call into a dependable service.
Abel Adamu Buba · 3 min read
Field notes
Technical notes on AI systems, RAG evaluation, latency and operational reliability by Abel Adamu Buba.
Evaluating RAG the way you would evaluate search.
Vibes are not an eval. A pragmatic harness for measuring retrieval quality, answer faithfulness, and regression risk across releases.
Abel Adamu Buba · 3 min read
Latency budgets and turn-taking in voice agents.
Sub-second turn latency, barge-in, tool calls, and safe handoff — the constraints that decide whether a voice agent is usable at all.
Abel Adamu Buba · 2 min read