Proof of Work

I build high-performance systems using Go, C++, and TypeScript. I've engineered a GFS-inspired storage system supporting 100M+ chunks (6.4PB capacity) and a C++20 sharded key-value store achieving 2.14M+ Ops/sec. My ML research on NOMA Cognitive IoT signal detection has been accepted in IEEE Wireless Communications Letters (IF: 5.4). Currently, I am contributing to Google Summer of Code '26 at P2P Labs, building resilience and backup infrastructure for Peersky Browser.


Education

Integrated B.Tech + M.Tech in Information Technology (CGPA: 8.18 / 10)
ABV-IIITM Gwalior | Aug 2022 — May 2027


Experience

Google Summer of Code '26 Contributor (May 2026 — Present)

P2P Labs | Peersky Browser

  • Building a Backup & Restore system for Peersky to preserve IPFS data, Hypercore feeds, chats, and browser sessions across failures.
  • Designing portable ZIP-based backups with CID-backed distribution over IPFS/Hypercore and cross-browser tab import support.
  • Improving PeerChat offline resilience via Hyperswarm UDP multicast discovery for seamless LAN messaging without internet.

Technologies: Electron, Helia, IPFS, Hypercore, Hyperswarm, TypeScript


Machine Learning Research Intern (May 2025 — Aug 2025)

ABV-IIITM Gwalior | IIITM

  • Developed an RBF-SVM signal detector for NOMA Cognitive IoT networks under high-noise channel conditions.
  • Engineered 10+ statistical and cyclostationary features, including cyclic autocorrelation, kurtosis, and skewness, with NumPy and SciPy.
  • Demonstrated a 3.6× throughput improvement over Cyclic OMA, reaching 0.6732 bps/Hz at 0 dB SNR.

Technologies: Python, NumPy, SciPy, scikit-learn, MATLAB


Software Development Intern (June 2024 — Aug 2024)

Maker Studios | Remote

  • Developed a secure Employee Training Tracker backend using Hono, PostgreSQL, and Drizzle ORM.
  • Implemented TOTP-based authentication and a sliding window rate limiter to mitigate unauthorized access.
  • Leveraged Effect.ts for functional, type-safe code, reducing runtime bugs by 90%.

Technologies: Hono, PostgreSQL, Drizzle ORM, Effect.ts, TOTP


Full Stack Web Developer Intern (Dec 2023 — Apr 2024)

DevTown | Remote | DevTown Platform

  • Engineered a high-performance React/Zustand learning platform, cutting initial bundle size by 30%.
  • Accelerated development by integrating 30+ REST APIs and architecting 20+ reusable UI components.
  • Resolved 10+ critical production bugs using Redux DevTools and advanced browser debugging.

Technologies: React, Zustand, Redux DevTools, RESTful APIs


Projects

DbSync: Distributed Key-Value Database (Jan 2026 — March 2026)

Live Demo | GitLab Repository

  • Engineered a C++20 shared-nothing key-value store achieving 2.14M+ Ops/sec with 34.05ms P99 latency, outperforming Redis by ~75% and DragonflyDB by ~84% in total throughput.
  • Implemented a multi-core asynchronous reactor using Linux io_uring and core-isolated shards to eliminate global locks and achieve near-linear scalability.
  • Architected an asynchronous batched Append-Only File (AOF) persistence engine, utilizing zero-copy RESP parsing and double-buffered io_uring flushes to decouple client responses from disk I/O.

MediRagAgent: Multi-Agent Medical Assistant (Jan 2026 — Feb 2026)

GitLab Repository

  • Architected confidence-based agent routing, agent handoffs, and human-in-the-loop physician verification.
  • Built multimodal Docling ingestion, dense+BM25 retrieval, query expansion, and TinyBERT reranking.
  • Deployed three PyTorch vision models with automated vision-based intent routing.

FraudLens: Cost-Sensitive Fraud Detection (2026)

GitLab Repository

  • Benchmarked five classifiers with class weighting, undersampling, and SMOTE using PR-AUC, calibration, precision, and recall.
  • Optimized decision thresholds for investigation costs and false-negative fraud losses.
  • Ran feature ablations and SHAP-based analysis of false-positive and false-negative cases.

ShellSync: Real-Time Collaborative Terminal (Oct 2025 — Jan 2026)

Live Demo | GitHub Repository

  • Built a collaborative terminal (Go, gRPC, React) streaming live shell I/O via bidirectional gRPC to a CLI agent, fanned out to web clients via WebSockets with low-latency resize sync.
  • Engineered multiplayer collaboration on a shared infinite canvas: live cursors, presence, and server-enforced role-based permissions (host / read-write/read-only) on password-gated sessions.
  • Added fault-tolerant state management with periodic Redis snapshots of serialized session state and graceful, leak-free shutdown of gRPC, HTTP, and background workers.

StudentSync: AI Academic Agent (Feb 2026 — April 2026)

Live Demo | GitHub Repository

  • Engineered a TypeScript AI student companion on Telegram and WhatsApp for coursework, notes, and habit tracking.
  • Designed a plugin-based LLM agent runtime with dynamic tool dispatch, task routing, and adaptive planning.
  • Automated Google Classroom workflows including assignment scraping, artifact generation, Drive upload, and submission.

CloudSync: Distributed Storage Cloud (Dec 2025 — Jan 2026)

GitHub Repository

  • Architected a Go-based, GFS-inspired distributed file system supporting 100M+ chunks (~6.4PB capacity) managed within an 8GB RAM control plane server.
  • Engineered a non-blocking Paced Replication Monitor to eliminate "stop-the-world" API pauses.
  • Designed a crash-consistent metadata engine using Write-Ahead Logging (WAL) and periodic checkpointing for data durability.

RuralSync: Hyperlocal Service Marketplace (Oct 2025 — Jan 2026)

Live Demo | GitHub Repository

  • Built a modular Node.js/MongoDB backend with role-based JWT authentication and reverse geocoding.
  • Integrated MongoDB Vector Search with geospatial radius queries for semantic service discovery.
  • Developed SEO-optimized city/service routes and automated sitemap generation (100+ pages).

Achievements & Competitive Programming

  • Amazon MLSS 2026: Selected among 3,000 of 150,000+ applicants
  • IEEE Wireless Communications Letters: Accepted Research Paper (Impact Factor: 5.4, 2025)
  • GATE 2026 Qualified: Computer Science
  • Flipkart GRiD 7.0: National Semi-Finalist, Software Development Track
  • LeetCode: Knight (1880+ Rating), Solved 1000+ Data Structures & Algorithms Problems
  • Codeforces: Specialist (1411 Rating)
  • CodeChef: 3-Star (1712 Rating; Global Rank 6 in Starters 243)
  • Smart India Hackathon: National Finalist (Top 50,000+ teams nationwide, 2024)

Technical Skills

  • Languages: C/C++, TypeScript, JavaScript, Go, Python
  • Backend & Databases: Node.js, gRPC, REST APIs, Express, Hono, FastAPI, PostgreSQL, MongoDB, Redis, AWS (EC2, S3)
  • AI/ML: PyTorch, scikit-learn, XGBoost, LightGBM, LangGraph, RAG, Qdrant, BM25
  • Frontend & Tools: React, Next.js, Redux, Zustand, Tailwind CSS, Docker, Git, BullMQ
  • Core Concepts: DSA, OOP, DBMS, Operating Systems, Computer Networks, Distributed Systems, Concurrency

Open Source Contributions

Peersky Browser (July 2025)

  • Bookmarks Feature: Implemented a complete bookmarks feature with favicon support and fallback handling.
  • QR Code Component: Built and refactored a QR code popup component with consistent styling and proper border radius.
  • UI/UX Fixes: Fixed URL bar hover state inconsistencies and optimized CSS.

Nightwatch.js (April 2024)

  • Element API Aliases: Added multiple alias methods (.attr(), .getCssValue(), .tagname(), .prop(), .text(), .accessibleName()) to the Element API.

Let's Connect

Ayush

Last updated by Ayush on September 12, 2026, 09:07 PM IST

ayushvish6555@gmail.com