Payments, queues,and query plans.

REST APIs and the services behind them: module boundaries, validation and DTO layers, versioning, and error responses a client can act on. Access control belongs at the boundary too — JWT and OAuth2 flows, role- and policy-based rules, rate limiting — so that individual features never have to reinvent it.

• NestJS·
• ASP.NET Core·
• Express·
• FastAPI·
• JWT·
• OAuth2·
• RBAC·
• PBAC

Schema design, indexing, and finding the query that is quietly costing the most. Most of the latency I have removed from a system came out of the data layer rather than the application code — a missing index, a relation loaded row by row, a table shaped for writing and then read in a completely different way.

• PostgreSQL·
• SQL Server·
• Redis·
• MongoDB·
• TypeORM·
• Prisma·
• Indexing

Queues, retries, distributed locks and scheduled work. Running a job is the easy part. The work is making sure it runs exactly as often as it should, that two workers never pick up the same unit of work, and that a failure overnight does not need a person awake to handle it.

• BullMQ·
• Redis·
• Idempotency·
• Distributed locking·
• Dynamic scheduling

Card charges, hosted payment links and refunds against the Tap gateway, and field service data through Zoho FSM. Integration work is mostly failure paths: signed webhooks, an audit trail you can reconstruct a dispute from, and retries that never charge someone twice.

• Tap·
• Webhooks·
• HMAC signing·
• Zoho FSM·
• Audit logging

I run code review for the backend team and plan sprints with our technical lead, at roughly 45 tasks per two-week cycle. Review is where risky migrations, N+1 queries and missing indexes get caught while they are still cheap. Generated code gets read more carefully than hand-written code, not less.

• Code review·
• CI/CD·
• Docker·
• Jest·
• Zoho Sprints

Running models locally where a network connection cannot be assumed, and building against hosted APIs where it can. The interesting constraints are usually not the model — they are latency, cost per call, and what the product does when the model returns something unusable.

• LangChain·
• Local model hosting·
• Gemini API·
• FastAPI