I'm Meet Sutariya, a computer science and data science undergraduate at Arizona State. I build across distributed transactions, storage and networking, resilient gateways, and applied retrieval — making failure boundaries explicit, then proving the guarantees hold.
Ira A. Fulton Schools of Engineering
B.S. Computer Science, with a minor in data science from the College of Liberal Arts and Sciences. Algorithms, systems and machine learning, pointed at distributed infrastructure.
The common thread is proof: make state and failure boundaries explicit, test the guarantees against real infrastructure, then measure before making a performance claim.
Enforcing content-bound scanner verdicts while keeping evaluation splits leakage-resistant and reproducible.
Tightening authentication races, runtime response validation, browser behavior, and frontend/backend contract checks.
Operating the retrieval service under explicit CPU, readiness, recovery, spend and infrastructure constraints.
Infrastructure automation for engineering labs and classroom systems.
Evaluation infrastructure for tool-using LLM agents.
Workflow orchestration for multi-step asynchronous jobs across distributed workers.
Four Spring Boot services own separate PostgreSQL databases and coordinate through Kafka. Transactional outboxes remove the database/broker dual write, idempotent consumers absorb redelivery, a persisted saga compensates partial failure, and trace context continues across HTTP, outbox relay and asynchronous consumers.
A persistent key-value server built below the framework line.
A versioned binary protocol feeds a bounded TCP server and sharded in-memory index backed by checksummed, append-only segments. The engine handles TTL expiry, compaction, truncated-tail recovery, corruption detection, and process-level ownership of the data directory.
A search and retrieval service measured under a fixed CPU budget.
Celery workers ingest and embed job postings into PostgreSQL; FastAPI combines GIN full-text search with pgvector HNSW retrieval through Reciprocal Rank Fusion. Redis coordinates queues, in-flight claims, answer caching and spend controls, while readiness gates model-backed routes until the per-process encoder is warm.
A resilient MCP gateway with content-bound admission enforcement.
The Go data plane handles protocol validation, legacy translation, safe retries, circuit breaking, bulkheads, backpressure, multi-tenant policy, SSE streaming, failover, metrics and tracing. Its admission gate reads content-hashed verdicts from PostgreSQL before routing; an out-of-band Python worker scans newly discovered tool descriptions, keeping a multi-second classifier off the request path.
A product with concurrency-safe auth and a typed API boundary.
A React/TypeScript client builds and tracks workout plans through a FastAPI/PostgreSQL backend. Refresh-token redemption is a single conditional database update, so concurrent tabs produce exactly one winner; zod validates responses at runtime and CI checks those schemas against the backend's generated OpenAPI contract. Plan generation retrieves from a ten-document knowledge base with BM25 and vector search fused by RRF.
Every project here is public, documented, and backed by CI. The benchmarks are local and labelled as such — none of them are presented as production capacity.
I’m interested in teams that care about system boundaries, measurable behavior, and software that survives more than the happy path — platform and cloud, backend, AI/ML, and general software engineering.