Software Engineer · Backend & Distributed Systems
M.S. Computer Science @ Vanderbilt University · graduating April 2027 · open to 2027 new-grad roles
I build backend services, storage systems, and developer tools. I like work where the numbers can be checked: every benchmark below links to the script that produced it.
| PyTorch contributor | Fix merged into core (#191092); all_reduce avg-reduction backward approved by a distributed maintainer (#190630). |
| Open-source maintainer | Reviewed and merged 112 pull requests from 4 contributors to lsm_tree and Proofdesk during ELUSOC 2026. |
| Publication | Co-author, An Integrated GCN–GAT–AE Framework for Robust Anomaly Detection in Industrial IoT Environments, IEEE CICT 2025. |
| Project | What it is | Evidence |
|---|---|---|
| LSM-Tree Storage Engine · C++17 | Collaborative key-value storage engine: lock-free SkipList MemTable, io_uring + O_DIRECT write-ahead log, blocked Bloom filters, leveled compaction. I lead and maintain it: architecture, design reviews, and integrating contributors' work. | 254K writes/sec, 32 µs P99 · benchmarks |
| Proofdesk · TypeScript, React | Collaborative textbook IDE: Y.js CRDT editing over authenticated WebSockets, BullMQ/Redis build queue, resource-capped Docker workers, in-browser Pyodide preview. | 358 ms median in-browser preview (local, 5 runs) · benchmarks |
| PulseStream · TypeScript, Kafka | Telemetry ingestion: Express → Kafka → PostgreSQL with device-keyed partitions, Redis idempotency keys, retries, and a dead-letter queue. | 3,991 req/s, 11 ms p50 / 34 ms p99 (local Docker) · load test |
| Satya · Python, Playwright | Agent that runs UI flows and flags success messages the backend contradicts (a "Deleted!" toast with no DELETE call). FastAPI service, SQLite run history, JUnit reports. | 77 tests in CI |
| ENGRAM · TypeScript | Memory engine for LLM agents: decay, reinforcement, contradiction detection, and hot/warm/cold tiered vector retrieval with int8 cold storage. | 73% fewer vector comparisons at equal Recall@5 (synthetic) · results |
| ReCL · PyTorch | Self-supervised learning for ultrasound/EEG sim-to-real transfer (InfoNCE + NCC reconstruction loss), with 3-seed baselines and ablations. | checkpoints · dataset |
- Amplify GenAI Innovation Center, Vanderbilt — Software Engineer, Backend & Platform Systems · Aug 2026 – present
FastAPI/Redis APIs for LLM tool calls and session state; pgvector hybrid retrieval that cut retrieval latency by 38%; QA that fixed 20+ defects. - VU-BEAM Lab, Vanderbilt — Research Assistant · Oct 2025 – Aug 2026
Built ReCL; async multi-GPU data loading that cut training time by 40%; Slurm sweeps on the ACCRE GPU cluster. - Mathematics Department, Vanderbilt — Software Engineer · Sep 2025 – May 2026
Built Proofdesk, a collaborative editor and sandboxed build system for PreTeXt textbooks.
Languages C++17 · Python · TypeScript/JavaScript · Java · Go · SQL
Backend & data Node.js/Express · FastAPI · PostgreSQL/pgvector · Redis · Kafka/Redpanda · BullMQ · SQLite
Systems & ML Linux (io_uring) · Docker · AWS · GitHub Actions · Prometheus · Playwright · React · PyTorch
Also: AWS Certified AI Practitioner · Microsoft Azure AI Fundamentals · Vanderbilt Global Good Hackathon finalist (2025, ARIA) · GDSC Technical Lead (2024–25)