Distributed systems · Agent infrastructure · Applied AI
I’m a Lead Backend Engineer at Hubtel and founder-engineer at Serendepify. I build stateful backend systems, agent infrastructure, and low-resource-language AI, with a bias toward the parts that fail in production: orchestration, concurrency, identity, deployment, evaluation, observability, and recovery.
Portfolio · Research · Writing · LinkedIn
| Project | The engineering problem | Explore |
|---|---|---|
| GroundControl | An agent-operable infrastructure control plane for applications on operator-owned servers. Remote agents use scoped MCP/OAuth actions instead of generic shell access; mutations become durable idempotent operations; connector capabilities are verified independently; exact deployed revisions can be reproduced in ephemeral sandboxes before repair. | Code + architecture · Host terminal design |
| RentAWeekend | A real-world agent execution system that turns uncertain conversational intent into research, saved choices, reviewed arrangements, payments, and human-executed work without silently crossing effect boundaries. Recent work includes row-lock/uniqueness protection for retries, transactional provenance, provider shadowing, and checkpointed recovery from model/provider failures. | Live product · Portfolio |
| Ghana Health AI | A Twi-first speech-and-language research stack connecting corpus provenance, ASR, human review, model adaptation, semantic evaluation, GPU training, and a deployed voice product. Model promotion is gated by evidence rather than demo quality. | Code · Research · Models |
| Backend as Natural Language | A Rust research runtime for lowering controlled natural-language backend declarations into typed intermediate representations, deterministic plans, validation traces, and executable capabilities. | Portfolio overview |
- Distributed and durable backends: C#/.NET, Temporal workflows, Kafka, Akka.NET, PostgreSQL, Redis, Elasticsearch/OpenSearch, failure handling, concurrency-sensitive financial and public-sector workflows.
- Agent infrastructure: scoped identity and authorization, long-running operations, idempotency, sandboxed reproduction, human approval boundaries, deployment/recovery tooling, MCP.
- ML systems and evaluation: Twi/Akan speech and meaning, corpus lineage, GPU training on Modal, model promotion gates, negative-result retention, and production inference.
- Developer infrastructure: Docker/Compose, Linux, Caddy/Nginx, GitHub Actions, Azure DevOps, cloud deployment, observability, and self-hosted control planes.
I’m interested in systems where the architecture has to survive retries, partial failure, unreliable model output, human approval, and production operations. I try to make the evidence inspectable: PRs, tests, evaluation records, deployment identities, benchmarks, failure traces, and explicit limitations.
My current research centers on low-resource Ghanaian language systems, especially Twi speech and semantic recovery. I keep research artifacts, reviewed data, model candidates, and production promotion as distinct evidence classes.
I also co-authored a peer-reviewed paper in African Geographical Review on geocoding polling-station data and accessibility in Ghana.
I’m especially interested in distributed infrastructure, agent/research platforms, ML systems, and teams building AI that has to operate reliably outside a notebook.
Explore the work · Read the engineering notes · Get in touch