Skip to content
View harsharajkumar-273's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report harsharajkumar-273

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
harsharajkumar-273/README.md

Hi there, I'm Harsha Raj Kumar! πŸ‘‹

Software Development Engineer (SDE) | MS CS Student at Vanderbilt University

I'm a Graduate Computer Science student at Vanderbilt University specializing in low-level systems programming, high-performance backends, and distributed systems.


πŸ› οΈ Technical Arsenal

C++ Java Python TypeScript Go
Node.js Express.js FastAPI React
Redis PostgreSQL Docker AWS

πŸ† Featured: Merged Core Contribution to PyTorch

pytorch/pytorch#191092 β€” fixed an uncaught ZeroDivisionError in torch.unravel_index, reviewed and merged by PyTorch core maintainers. Six additional PRs across distributed, linalg, and BLAS code are currently in review.

πŸ“„ Publication

An Integrated GCN–GAT–AE Framework for Robust Anomaly Detection in Industrial IoT Environments β€” co-authored with Suhani Panda, Sanjana B, and Saravanan Palani. Published in the 2025 IEEE 9th International Conference on Information and Communication Technology (CICT), Chennai, India (Dec 2025). A seven-stage hybrid GCN/GAT/Autoencoder framework for real-time IIoT anomaly detection, achieving 95.0% classification accuracy and 8.1ms inference latency on the Edge-IIoTset dataset.


πŸ’Ό Professional Experience

  • Amplify GenAI Innovation (AGI) Center, Vanderbilt University | Software Engineer β€” GenAI Platform & Systems (Aug 2026 – Present)
    • Engineering scalable GenAI platform tooling, API integrations, and multi-agent workflows across Vanderbilt's College of Connected Computing.
    • Building high-throughput LLM middleware services and vector indexing pipelines to accelerate AI application deployment.
  • VU-BEAM Lab, Vanderbilt University | Research Assistant (Oct 2025 – Aug 2026)
    • Designed ReCL (Reconstructive Contrastive Learning), a self-supervised ultrasound model evaluated against baselines on the PICMUS benchmark. Submitted to NeurIPS 2026 (not accepted).
    • Optimized offline batch-processing data pipelines across 10,000+ ultrasound scan frames, reducing multi-GPU training execution time by 40%.
  • Vanderbilt University Mathematics Department | Platform Architect & Software Engineer (Jan – May 2026)
    • Engineered Proofdesk, a collaborative Web IDE sandbox, reducing compilation feedback latency by 72% (from 1.1s to 300ms) via client-side WebAssembly (Pyodide).
    • Deployed WebSocket sandboxed terminal runtimes (node-pty) inside resource-restricted Docker containers (512MB RAM limit).
    • Architected a distributed background worker task queue using Redis and BullMQ with local fallback loops.

πŸŽ“ Academic Credentials

  • 🏫 Vanderbilt University (Nashville, TN) | Master of Science in Computer Science (Expected Jun 2027)
  • πŸŽ“ Vellore Institute of Technology (VIT) (Chennai, India) | Bachelor of Technology in Computer Science (Graduated May 2025)

⚑ Flagship Systems & Implementations

πŸ“ˆ PulseStream Telemetry Pipeline (TypeScript, Redpanda/Kafka, KEDA)

  • Designed a horizontally scalable telemetry ingestion gateway targeting 50,000+ metrics/sec across distributed Redpanda (Kafka-compatible) event streams (see repo for benchmark-reproduction script).
  • Decoupled write paths from persistence using device ID hash partition keys for fast HTTP 202 ingestion ACKs.
  • Implemented double-layer idempotency utilizing Redis SETNX edge locks and atomic PostgreSQL batch upserts.

πŸ” revert-ai (TypeScript, Node.js, AST Parsers, Git API, NPM)

  • Built and published revert-ai on npm β€” a dependency-aware undo/redo CLI for AI coding assistants (Claude Code, Aider), reading session logs and git history to track file operations.
  • Engineered AST-based dependency analysis for JS/TS and Python import graphs to safely cascade file reversions while preserving unrelated edits.
  • Implemented side-by-side terminal diff previews and Git stashing safety guards, validated by a Jest test suite.

πŸ’Ύ LSM-Tree Key-Value Engine (C++20, io_uring)

  • Engineered a high-performance LSM-Tree database storage engine in C++20 with a small team, via a structured open-source-club build (issue-driven PRs, CI, dependabot).
  • Implemented asynchronous, non-blocking Write-Ahead Log (WAL) appends utilizing Linux io_uring with O_DIRECT, bypassing the kernel page cache.
  • Engineered a concurrent, lock-free SkipList MemTable with atomic CAS pointers and a custom bump-pointer memory Arena.
  • Restricted read lookup penalties to at most one CPU cache line miss using 64-byte cache-aligned vectorized Block Bloom filters.

πŸ… Recognition & Leadership

  • Vanderbilt Global Good Hackathon Finalist (2025) β€” co-developed ARIA Pathfinder, a PostGIS-backed disaster response routing system using Dijkstra with time-decaying edge weight equations.
  • GDSC Technical Lead (2024–2025) β€” co-led the Technical Division for Google Developer Student Club (GDSC), running technical workshops for 200+ student members.

πŸ“Š GitHub Contribution Stats

Harsha's GitHub Streak

Harsha's Trophies

πŸ”— Connect with Me

Pinned Loading

  1. API-gateway API-gateway Public

    Production-grade API gateway β€” Redis sliding-window rate limiting, JWT auth, circuit breakers

    JavaScript

  2. Netflix-RL Netflix-RL Public

    An AI-driven, asynchronous load balancer prototype using Reinforcement Learning (Linear Thompson Sampling Contextual Bandits) and FastAPI to mitigate massive thundering-herd traffic spikes under su…

    Python

  3. Proofdesk Proofdesk Public

    Proofdesk is a high-performance web-based IDE and compilation sandbox designed for collaborative authoring, testing, and rendering of mathematical textbooks.

    TypeScript 4 3

  4. PulseStream PulseStream Public

    A high-throughput, real-time IoT metrics ingestion pipeline implementing advanced distributed systems patterns: API key caching, edge idempotency locking (Redis), partition-key event streaming (Red…

    TypeScript

  5. Repost-Radar Repost-Radar Public

    A hyper-optimized, ultra-low-latency real-time text deduplication engine using C++20, AVX2 SIMD vectorization, MinHash, and Locality-Sensitive Hashing (LSH) for sub-millisecond duplicate checks on …

    JavaScript