Vedansh Adepu

I’m a database-systems software engineer with 2.5 years building reliable, low-latency backend and data infrastructure. I’ve reduced p99 latency from 150ms to 18ms through lock-free concurrency design, and I’ve operated a 1.3B-events/day ingestion tier at 99.9% SLO on AWS. I care deeply about correctness under failure, measurable performance, and data quality that holds up in production.

At SciEncephalon, I built a checkpoint/resume framework for multi-stage data pipelines with typed schema contracts, deterministic replay, and automated end-to-end validation—raising automated coverage from 0% to 85% while cutting manual intervention 65%. I instrumented microservices with OpenTelemetry distributed tracing to isolate and remove a concurrency bottleneck in the token-streaming layer, improving throughput and lowering P99 latency under identical load. I also integrated a FAISS vector index into a RAG pipeline with retrieval-quality evaluations (recall@k and rubric-based review), cutting off-context answers by 22% versus the baseline.

Before that, at KuppiSmart Solutions, I architected a gRPC prediction API using Spring Boot and Protobuf, serving 100k+ daily requests with service-layer overhead under 50ms and reducing schema-break incidents 70% via CI/CD schema-diff gates. At Brane Group, I replaced a mutex queue producer serialization with a lock-free MPSC ring buffer and redesigned an ingest service using an epoll-based non-blocking I/O reactor to sustain 8,000+ concurrent connections. My professional ethos is simple: design for failure, measure everything (including latency and recall), and turn production lessons into durable systems like runbooks and replayable workflows.