Harsh Gupta

Harsh Gupta

21 · AI Engineer

I love building, breaking, and shipping intelligent, production-grade applications. Currently contributing to open source as a GSoC '26 contributor at Joplin, and expanding my knowledge in Machine Learning & Agentic workflows. Always open to collaborate, chat, or geek out over cool tech stuff.

Experiences

Google Summer of Code 2026
Google Summer of Code 2026
Joplin
May 2026 - Present
Remote
Show details
  • Building an on-device note-categorization plugin using Transformers.js/ONNX, keeping all embedding computation local.
  • Designed a chunk-embed-average pipeline with cosine-similarity-based title weighting, selecting the BGE-small-en-v1.5 embedding model after benchmarking clustering accuracy, memory footprint, and inference speed.
  • Implemented semantic clustering with UMAP dimensionality reduction and multiple strategies (HDBSCAN, K-Medoids, K-Means), achieving a top clustering score of 0.94 across 7 auto-detected topic clusters.
  • Built a tag-suggestion pipeline combining TF-IDF keyword extraction with centroid cosine-similarity re-ranking to generate descriptive multi-word tags for each cluster.
Transformers.js
ONNX Runtime
Python
LangChain
Open Source Contributor
Open Source Contributor
Joplin, AsyncApi & more
Dec 2025 - Jun 2026
Remote
Show details
  • Joplin: Authored 12 pull requests, with 11 merged, resolving long-standing issues across the Desktop application, Markdown editor, rendering engine, exports, and localization.
  • Joplin: Fixed the Custom Dictionary storage implementation by redesigning how Custom Dictionary.txt is located and persisted across platforms, resolving a long-standing issue after evaluating multiple implementation approaches.
  • Joplin: Improved the Markdown editor by fixing whitespace formatting edge cases, fenced code block completion, and clipboard rendering from the Markdown preview, making editing behavior more consistent.
  • Joplin: Enhanced the desktop experience by implementing auto-scroll from Go to Anything, improving ABC Sheet Music rendering, enriching Markdown exports, and adding customization for internal note link icons.
  • AsyncAPI Generator: Authored 18 merged pull requests, substantially expanding automated test coverage for the JavaScript WebSocket Generator and improving project reliability.
  • AsyncAPI Generator: Improved developer tooling through deterministic Turborepo caching, AsyncAPI v3 documentation updates, WebSocket error handling improvements, and CI consistency.
React
Node.js
TypeScript
Full Stack Engineer
Full Stack Engineer
Intern, GRS Worker
Sep 2025 - Nov 2025
Delhi, India · Remote
Show details
  • Architected full-stack app with Next.js 15, Prisma, NextAuth; secured sessions and complex DB relations.
  • Integrated Twilio SMS, QR codes, Zod validation; cut data errors by 40%.
  • Built UI with Radix UI, Framer Motion; optimized API with Turbopack and Server Actions.
Next.js
Prisma
TypeScript
React
Node.js

Projects

I love designing and building thoughtful, production-grade applications.

View Project

botforweb preview

botforweb

Your docs, turned into a fact-checked AI assistant. Upload PDFs and Markdown to get a citation-grounded chatbot live on your site in under 5 minutes.

All Systems Operational

View Project

Chess preview

Chess

Real-time multiplayer chess platform featuring live in-game chat, timed 5-min games, instant matchmaking, and puzzles.

All Systems Operational

Skills

I love working with these technologies to build beautiful, functional, and intelligent applications.

Applied AI & Agents

LangChain
LangGraph
RAG Pipelines
Agentic Workflows
Fine-Tuning & LoRA
LLM Evaluation (Ragas)
Vector Search & pgvector
DSPy
Semantic Caching
Structured Outputs

Core Development

Python
FastAPI
TypeScript
JavaScript
React
Next.js
Node.js
Express

DevOps & Infrastructure

PostgreSQL
Prisma
Docker
Kubernetes
AWS
Git
C
Rule 7: Fights will go on as long as they have to.