About
Gamaliel Dashua
Backend / AI Engineer · open to remote work
I spend most of my time on the part of a system that decides whether anyone can tell what went wrong. That interest is not academic: I have been the person combing through logs across services for the better part of a day, looking for one bad run out of thousands, while every dashboard insisted everything was fine.
Before that, backend work on LAMISPlus — the electronic medical record Nigeria's national HIV/AIDS programme runs on, where a duplicated record splits somebody's treatment history and a silently dropped visit means the wrong regimen at the next appointment. That is where the habit of validating at the boundary came from, rather than from a style guide.
The capstone on this site is that habit, applied to AI agent pipelines: one trace ID that reaches every step, and an authoritative watcher that sits outside the process it is watching, so it can still speak after that process is gone.
How I work
Four things that show up in every repo
Right-sized tools
The impressive-sounding choice is usually the wrong one. A handful of pages of text and images does not need a framework behind it; the capstone runs Redis and Docker because it genuinely needs a process boundary to argue across.
Failures that do not stay silent
A clean exit is necessary and not sufficient. I would rather a build break loudly at the boundary than a wrong number reach someone who trusts it.
Legible history
Atomic commits, one logical change each, and a tag at every milestone.
Written down
If I cannot explain why a tool is in the stack, it comes out. Every choice on this site has a reason I can say out loud.
Stack
What I reach for
- Backend
- Java · Spring Boot · Python · FastAPI · PostgreSQL · Redis
- AI pipelines
- LangGraph · LangChain · LangSmith
- Front-end
- React · TypeScript · Astro
- Infrastructure
- Docker · Cloudflare Pages · GitHub Actions
Current focus
Open to remote work
Looking for backend and AI-pipeline work with a team that runs agents in production — the kind where a run failing quietly is somebody's actual problem on a Monday morning. Remote, and comfortable across timezones; the Logbookie contract was exactly that.