YakhaLabs builds that software: applications, automations, AI systems, and analysis pipelines for small businesses and research teams.
The problem
The same numbers live in three systems, and nobody trusts any of them.
Everyone says AI should be saving you time. Nobody can point to where.
The data took days to clean, and the figure in the draft still can’t be traced back to its source.
None of that is a people problem. It’s tooling, and tooling can be built.
Services
When the tool you need doesn’t exist: web apps, internal tools, dashboards. Scoped small enough to ship, documented well enough to own.
When you know AI should help, but not where: we find the job worth testing first, build it into the tools you already use, and measure whether it paid for itself. Under the hood: LLM integrations, agentic pipelines, RAG knowledge bases, evaluation, rollout.
When the week disappears into manual steps: we map how work actually moves through your business, find the drag, and remove it with system integrations, scripted workflows, and AI-assisted steps. Measured in hours saved and errors that stop happening.
When you need answers you can defend: analysis pipelines, dashboards, prediction models, experiment design, and figures fit for a journal reviewer.
The plan
A short call or an email. Plain words are enough; you don’t need a spec, and you won’t get jargon back.
What gets built, what it costs, and what it should change: hours saved, errors gone, questions answered. Small first pieces, not big-bang projects.
Tested, documented, and yours. Before handoff, we check the result against the numbers from step 2.
Proof
YakhaLabs is new, and it doesn’t arrive empty-handed. Recent builds:
Why does this line of code exist? wipnote answers that question six months later. It links commits, AI agent sessions, and work items into a causal chain you can walk in either direction.
Go · SQLite · HTML as the data store
A machine-learning model that predicts World Cup matches, scored in public against every result. The shows on pitchcasts walk through the hits and the misses. Choose your favorites and see if you beat it.
Python · match-prediction model · pitchcasts.com
Thousands of charging-station reviews, scraped and run through an LLM that pulls out the recurring themes. A dashboard on Parquet files shows what drivers actually complain about, counted.
scraping · LLM theme extraction · Parquet
Reproducible research pipelines on where AI’s capital goes and what the grid can carry. Every figure is publication-grade, and every number traces back to its query.
marimo · DuckDB · dlt
A multi-agent Claude Code workflow that drafts long-form documents from a corpus of verified facts. Nothing ships until every claim passes a fact-check gate.
Claude Code · multi-agent workflow
Browser automation that turns dense source material into customized NotebookLM audio overviews: prompt generation, source selection, and coverage checks against the finished transcript.
Claude Code · browser automation · NotebookLM
About
I’m Shakes, the builder behind YakhaLabs. I’ve watched good teams lose whole weeks to work a script could do, and I spent my analytics career fixing that in energy, on solar portfolios half a million systems deep. The tools I kept building there are the reason this company exists.
In siSwati, yakha. In my wife’s Swahili, jenga. Either way: build.
After
The report writes itself before Monday’s meeting.
The numbers agree with each other, and with reality.
When someone asks whether the AI paid off, you can show them.
Contact
Tell us what you’re trying to build, automate, or figure out. You’ll get a straight answer on whether YakhaLabs fits, and what it would take.
Or skip the form
hello@yakhalabs.com