Work with a reason.
Systems with a story.

A declarative, reproducible plugin manager and curated distribution for Herdr.

Challenge

Keep a Herdr setup reproducible across machines without turning plugin configuration into a hand-maintained pile.

Approach

Model plugins as a plain list, lock exact commits, and provide a CLI/TUI workflow for syncing a setup.

A bug-fix contribution to Headroom, an open-source context compression layer for LLM applications.

Challenge

Docker deployments could ignore the configured port, leaving health checks and diagnostics probing the wrong port.

Approach

Updated container startup and health checks to honor HEADROOM_PORT, and taught headroom doctor to resolve the configured deployment port.

A focused bug-fix contribution to TiDB, a Go-based distributed SQL database by PingCAP.

Challenge

Turn a concrete optimizer-hint warning issue into a small, reviewable open-source contribution.

Approach

Followed the issue from a good first issue through implementation, two LGTM reviews, and passing CI.

Built a BFF that delivers daily operating profit data to a management dashboard while migrating the source from Blob to BigQuery.

Challenge

Give a management dashboard a stable API while the reporting data source and its access model were changing.

Approach

Worked across API design, OpenAPI contracts, authentication, authorization, testing, and the Blob-to-BigQuery migration.

A near-real-time pipeline processing roughly 5 billion parcel records per year at ten-minute intervals.

Challenge

Replace a legacy flow while keeping large-scale parcel processing reliable and incremental.

Approach

Rebuilt the pipeline around idempotent incremental processing and parcel-ID deduplication.

Built a platform that consolidates company-wide management information into a single source of truth.

Challenge

Bring fragmented management information together while supporting layered permissions and dynamic access control.

Approach

Designed the web application and authorization model, then migrated analytics from Power BI to the new interface.

Evaluated whether GraphRAG is necessary for a real-world use case by comparing it with a standard RAG baseline.

Challenge

Decide whether the added complexity of GraphRAG was justified for the use case.

Approach

Compared both approaches across answer quality, cost, and latency to make the trade-offs visible.