Data engineer in Nairobi. I build data platforms for civil-liberties research, and the checks that make their numbers trustworthy.
CLIO (Civil Liberties Intelligence Observatory) is a method for turning heterogeneous information-control signals into validated, uncertainty-stated evidence. It is piloted in Kenya and built to extend to other countries. It joins network-interference measurements, conflict events and platform transparency data into governed data marts, and classifies every week into a pressure regime.
Start here: the CLIO live dashboard and the CLIO Findings Explorer.
flowchart LR
OONI["OONI network measurements"] --> P["Bruin pipelines on BigQuery"]
ACLED["ACLED conflict events"] --> P
GTR["Google Transparency Report"] --> P
P --> M["Weekly classified data marts"]
M -- "committed snapshot" --> CFE["CLIO Findings Explorer"]
| Project | What it shows |
|---|---|
| CLIO Findings Explorer · live | Five Kenya findings, each one classified week with its evidence and a limitations box. 671 tests, including one that recomputes all 2,029 numbers on the evidence cards from the raw snapshot. Cloud Run and Neon, auto-deploy to dev, manual promotion to production, OpenTelemetry and an on-call agent with usage caps and a kill switch. |
| CLIO | The platform itself: ingestion, SQL data marts and the weekly regime classification. |
| Civil Liberties Knowledge Assistant | A RAG assistant on internet censorship and digital rights in East Africa (2022 to 2026), grounded in OONI, Access Now, CIPESA and Freedom House sources. Every answer cites its source. |
| agent-relay | An HTTP task relay extended with OpenTelemetry, a per-route 5xx alert and a read-only incident responder gated by code policy and human approval. |
Agents write much of my code. I put the effort into the checks around them: a git hook that stops agents from committing, a spec-first guard on edits, tests that recompute every published number, and a public log of what each agent did and what a human decided.
BigQuery · SQL · Python · Bruin · Terraform · FastAPI · PostgreSQL · Docker · GitHub Actions · OpenTelemetry · Grafana
Answers I've contributed to the DataTalksClub course FAQ: #497, #422, #409, #314.
I write about what I build and what went wrong on LinkedIn and X.


