SYSTEMS THAT KEEP WORK MOVING SYSTEMS THAT KEEP WORK MOVING

DATA THAT STAYS CORRECT UNDER FAILURE DATA THAT STAYS CORRECT UNDER FAILURE

IDEMPOTENT JOBS THAT RETRY SAFELY IDEMPOTENT JOBS THAT RETRY SAFELY

QUERIES SHAPED FOR THE DATA THEY READ QUERIES SHAPED FOR THE DATA THEY READ

POINT-IN-TIME RECOVERY BEFORE INCIDENTS POINT-IN-TIME RECOVERY BEFORE INCIDENTS

DISTRIBUTED TRACING AT SERVICE BOUNDARIES DISTRIBUTED TRACING AT SERVICE BOUNDARIES

Loading

t.sabudh
About, Blog, Contact

Backend / platform / reliable data systems

I build backend systems where data stays correct under load, retries, and failure.

I am a backend engineer in Kathmandu, Nepal. I build backend platforms, event-driven workers, query paths, and observability surfaces for systems where duplicate work, delayed jobs, data drift, and partial failure have to be handled correctly.

At Relativity Invest I own the backend and data infrastructure for a production financial-data platform: the schemas, the workers, the recovery paths, and the calls on what is safe to ship. Backend work since 2022.

Event-driven architecture Record linkage Query optimization Production recovery Observability

Engineering profile

Backend-heavy work, measured by failure behavior.

My strongest work is in systems where the hard part is not drawing the interface, but keeping data correct while load, retries, schema changes, and partial failure are happening at the same time.

The systems I want to keep building are change data capture pipelines, event-driven workers, analytical query paths, database performance work, data reconciliation, precision-critical components, and observability surfaces that prove the system is healthy.

I design service boundaries around clear contracts, idempotent job processing, bounded queues, recovery paths, and database queries that match how the storage engine actually reads data.

Accomplishments

Query optimization for stock screeners

Reworked slow stock-screener workloads so large result sets moved from 8-22 seconds to under 600ms, using read-model separation to keep transactional database paths isolated from analytical reads.

Idempotent notification processing

Made notification workers safe under duplicate messages, retries, failed deliveries, graceful shutdowns, delayed processing, and replay from dead-letter queues.

Multi-source entity resolution

Linked the same executive arriving from many sources under different identifiers, rebuilt canonical records nightly across about 6,000 companies with weighted source precedence, and kept append-only column-level provenance so a disputed field traces to its sources and a bad source can be retracted.

Point-in-time recovery and migration safety

Recovered deleted production rows from a point-in-time database branch, then added migration safety checks that block known destructive SQL patterns before deployment.

Project lanes

Backend platforms

APIs, workers, auth, service boundaries, and production recovery.

Data infrastructure

Query optimization, read-model separation, audit trails, and query routing.

Distributed systems

Event-driven workers, idempotent processing, bounded queues, replay, ordering, and lag visibility.

Systems components

Precision-critical components, protocol-adjacent tooling, and latency-sensitive code.

Home
About
Blog
Resume
Hire
tsabudh@gmail.com
Github
Linkedin
Medium

Sabudh Thapa, backend engineer in Kathmandu, Nepal.