OYINBO, David Bayode

Sole architect, engineer, and operator of a public-beta, multi-region real-time platform spanning backend services, frontend, cloud infrastructure, deployment, observability, documentation, and SDK-facing work.

• Own the end-to-end design and implementation, including database design, Rust backend services, the Nuxt/Vue dashboard, AWS infrastructure, deployment, observability, documentation, and SDKs.
• Built a RESP-inspired, zero-copy Rust WebSocket protocol. In recorded Criterion tests, its parser stayed near 0.3 microseconds from 128 B to 256 KiB; at 256 KiB it took 305 nanoseconds versus about 38 microseconds for serde_json, roughly 125 times faster in that test.
• Used Kafka, Redis, ClickHouse, and HAProxy for per-channel ordering, multi-node fan-out, usage history, and selective cross-region replication.
• Provisioned US and EU infrastructure with Terraform, Kubernetes, Amazon EKS, MSK, EC2, ElastiCache, and Route 53, with OpenTelemetry and the Grafana observability stack.
• Sustained 96,114 messages per second for 25 minutes with 0.5 KiB payloads on one c8g.large (2 vCPUs, 4 GiB memory), using 102 MiB memory and 56% CPU at 32 ms p95 round trip including k6 overhead. The load generator failed before server saturation, so this remains a tested operating point.
• Created and independently maintain Trypema, distkit, and Laye for reusable rate limiting, distributed counters and locks, and authorization policies.

Owned a multitenant accounting service from design through production delivery in a team of about 10 engineers, reporting directly to the CTO. Also rebuilt a resource-heavy automatic bet placement service and contributed to shared APIs, observability, email, and dashboard work.

• Designed the NestJS and TypeScript accounting service with PostgreSQL, Drizzle ORM, atomic transactions, tenant isolation, relational constraints, and immutable audit trails.
• Built 794 passing automated tests across 35 Vitest and Supertest suites, executing in under 30 seconds as of Jan. 2026.
• Rebuilt the Python/FastAPI automatic bet placement service with caching, Dockerized deployment, unified Vertex and Pinnacle adapters, and end-to-end observability.
• Reduced p95 latency from 3,555 ms to 400 ms, removed an 8.7% failure rate in the comparison workload, and increased throughput from 9 to 45 requests per second per node.
• Collaborated on a TypeScript-to-Go migration of betting-platform APIs: endpoints, SMTP, OpenTelemetry/Grafana.
• Built an account-linked RAG assistant across WhatsApp, Telegram, and Slack using LangChain, ChromaDB, PostgreSQL, custom APIs, and OpenAI/Anthropic models; a blinded 30-question evaluation reached 0.843 judged relevance at 13 times lower cost than direct GPT-5.5.