Backend / AI Engineer · Available for remote work

I build observable AI agent pipelines you can trace end‑to‑end — every step back to one ID.

When a run fails, one ID links the failure to the exact step, prompt and model call that caused it — including the failures that normally leave no trace at all.

See the trace ↓

Gamaliel Dashua, Backend / AI Engineer

Lead case · LangGraph agent pipeline observability

Time to root cause: ~14 hrs → under 2 min

A run fails and every dashboard stays green, because the code that reports the failure was running inside the process that died. Move the authoritative watcher one level up and the same failure is one paste away.

trace ID: 4161ab9f-17f7-4ced-a552-2ba12de4e6cc

The trace page for one run. A navy trace ID with an amber marker beside it, a banner reading "The watchers disagree", a quarantine notice, and two cards — LangSmith reporting "finished" and the host observer reporting "killed_out_of_memory" with exit code 137 and the exact script that ran.
GET /trace/{id}. The disagreement is the demo: one watcher says fine, the other says dead and here is the body — under a single ID. LangSmith is not lying. The agent planned, wrote the code and handed it off, and all of that genuinely worked. The kill happened afterwards, somewhere it was never present.

Read the case study →

In production

Shipped and maintained LAMISPlus, the national HIV/AIDS electronic medical record deployed across Nigerian treatment facilities.

  • Java / Spring
  • FastAPI
  • LangGraph
  • React / TypeScript

Selected work

Three that show the range

LangGraph agent pipeline observability

One trace ID that survives the crash LangSmith cannot see, because the authoritative watcher sits outside the process that dies.

headline metric: root cause — ~14 hrs → under 2 min

LAMISPlus — national HIV/AIDS EMR

Backend work on the electronic medical record Nigeria's national HIV/AIDS programme runs on.

headline metric: production at national scale

Logbookie.eu

React and TypeScript front-end work delivered remotely, on contract, for a European team.

headline metric: remote contract, European team

All 5 cases →

About

Backend and AI engineer. I have shipped a national-scale medical record system and spent long enough debugging pipelines in the dark to build the thing that stops it happening twice.

More about me →

Get in touch

Send me an email

I would rather be judged on the work than on a CV, so a practical exercise suits me just as well as a conversation.

Get in touch →