I build and operate backend systems for SaaS and civic-tech platforms — APIs, distributed pipelines, AI/LLM infrastructure, and the observability that keeps them running when nobody's watching.
# Most recently a Senior Software Engineer at Search Atlas, an agentic SEO platform — backend services integrating Google Search Console, Analytics, Ads, and Meta Ads. Previously three years at Code for Africa building civic-tech data platforms used by researchers and journalists across the continent.
# Recent work centers on applied AI systems: gated RAG pipelines, LLM routing with automatic failover, and self-hosted model platforms where data sovereignty matters. AWS Solutions Architect certified. Open-source contributor to Django and CKAN.
Backend services for an agentic SEO platform: DRF APIs and Celery workers over Google Search Console, Analytics, Ads and Meta Ads. Extended the LLM orchestration layer with multi-provider routing, automatic failover, and cost-aware model selection. Fixed a Celery Beat race that sent customers duplicate report emails, using Redis-backed distributed locks. Instrumented Django and Celery with OpenTelemetry, and tuned PostgreSQL and ClickHouse for high-volume analytics tables. Claude Code was the daily workflow, from issue-driven plans through AI-assisted review in CI.
Backend and infrastructure for civic-tech and open-data platforms used by researchers, journalists, and policy teams across Africa. Lead maintainer of openAFRICA, the CKAN-based open-data portal, through upgrades from CKAN 2.8 to 2.11. Led the architecture and Ansible-driven deployment of multi-node Elasticsearch clusters, gated on a Rally benchmark harness before production traffic. Built end-to-end observability on Grafana, Prometheus, Loki, and Alertmanager, and drove CI/CD adoption across engineering teams.
Tendersure Africa, a B2B procurement SaaS: Django and Vue.js in English, French, Portuguese and Amharic, with localised payment gateways, signed callbacks, and idempotent retries; Terraform provisioning, VM-to-ECS container migration, CloudWatch and Prometheus alerting.
Python and Flask data applications over PostgreSQL, dbt-based ETL pipelines, and JupyterHub deployed on Docker Swarm for the analytics team.
First engineer on the product: built jobsearchke.com from scratch in Django, PostgreSQL, and JavaScript, deployed on DigitalOcean behind Docker and Nginx.
Led development of an internal ad-quality classifier built as a gated multi-stage RAG pipeline — cheap signals evaluated first, so expensive model calls run only when they change the outcome.
Routing layer across leading LLM providers with automatic failover and cost-aware model selection — one interface for product teams, resilience and spend control underneath.
Designed and deployed an 8-node Elasticsearch cluster — 3 master-eligible, 5 data nodes — with custom Ansible roles for re-indexing, ILM, and snapshot lifecycle management. Benchmarked with Elastic Rally.
Led the deployment of a self-hosted LLM platform to complement commercial providers — a private option where data sovereignty and predictable inference costs mattered most.
Lead maintainer of Code for Africa's CKAN-based open-data portal and its custom extensions. Carried it through CKAN upgrades from 2.8 to 2.11, including the move to Python 3, and migrated its legacy controllers to Flask blueprints.
Contributed to a distributed news ingestion pipeline operated by Mediacloud, a nonprofit research collaboration — an open archive used by researchers studying media ecosystems globally.
Backend services for an IoT air-quality network across multiple African cities — geo-tagged ingestion and async aggregation tolerant of intermittent connectivity.
Census and survey visualization platform with nested administrative geographies, tuned for spatial queries serving interactive maps across multiple African countries.
Live XAU/USD trading bot in Python on MetaTrader 5, running several strategy configurations in parallel with Telegram trade notifications. Every strategy change is validated in a custom backtester against historical data before going live, with position and exposure limits enforced in code.