~/home → AI work

AI work & case notes

Practical AI, not noise. The focus is agents, RAG, evaluation, networking, MikroTik, ISP billing, cloud computing and the kind of full-stack engineering that turns experiments into systems people can use.

flagship direction · practical AI 01/05 agents · RAG · evals

Practical AI systems for real workflows

My center of gravity is AI that can be deployed, measured and trusted: assistants that retrieve grounded context, agents that call tools safely, workflows that reduce manual work and scorecards that show whether the model is actually useful.

THE PROBLEM

Too many AI demos stop at a pretty answer. Real teams need systems that know their context, control model costs, expose failure modes and fit the workflow instead of interrupting it.

THE APPROACH

Build from the workflow backwards: data sources, retrieval, prompts, tools, UI, evaluation and deployment. Useful AI is an operating system for work, not only a chat box.

  • Agents with tools — task flows that can search, update records, summarize context and hand off when confidence is low.
  • RAG with grounding — retrieval pipelines that keep answers tied to source material, private knowledge and local context.
  • Evaluation loops — measure cost, latency, failure rate, hallucination risk and whether the output helped the user finish the task.
  • Production delivery — APIs, dashboards, auth, storage, queues, monitoring and cloud deployment around the model.
Follow the AI notes GitHub profile
llm APIs · agents · RAG · evals · SaaS · cloud
// focused builds

Where the AI work is going

02/05 — 05/05
case 02

Hakili AI — legal intelligence direction

Kenyan legal research should become easier to navigate without becoming reckless. The direction is a grounded assistant: retrieve the right law, cite the source, explain the reasoning and show uncertainty clearly.

legal AIRAGcitationssearch UXevaluation
trust-first · source-grounded build notes
case 03

MUA Jobs — AI-assisted employment

Job search needs matching plus guidance. The direction is AI support for fit analysis, application improvement, skills gaps and opportunity discovery, tuned for the way people actually look for work in Kenya.

matchingcareer agentsSaaSuser workflowsAfrica
people-first · practical guidance progress thread
case 04

AI implementation scorecards

I care about the boring numbers that tell the truth: local developers trained, Kenyan datasets governed responsibly, pilots moved into production and teams able to audit failures when models break.

evalsmodel qualitypolicycostsadoption
measure what matters on X
case 05

Software depth behind the AI

The AI work sits on real engineering: networking, MikroTik, ISP billing, databases, APIs, dashboards, payment flows, server automation and production operations. That background keeps the model layer attached to systems that can run every day.

networkingMikroTikISP billingcloudops
deployment-minded AI GitHub

Building AI that should move from idea to production? Let’s talk.

X — @tim_maundu