Yash Gangurde

AI/ML Engineer | Pune, India

Final-year IT Engineering student building production AI systems, LLM pipelines, and backend automation.

Projects:

Experience:

Contact via Email
GitHub Profile
LinkedIn

I BUILD
THINGS
THAT THINK

Not a tutorial follower. Shipped a multi-character anime AI with a custom LLM fallback chain. PDF tool with 5K+ organic users. Automation agents that run in production.

PROJECTS

A.R.C.A.N.E.

Flask, Groq, Gemini, Ollama, Tavily, PWA, JWT

• Engineered a fault-tolerant 3-tier LLM fallback chain (Ollama → Groq → Gemini), ensuring zero downtime and stateful context.

• Built 15 distinct anime character pipelines handling high-volume inference with dynamic temperature and context injection.

• Integrated Tavily search via Groq tool calling for real-time contextual responses.

• Hardened backend architecture against LLM jailbreaks using custom regex heuristics and strict JWT-based session isolation.

View Live →

PDF360

Flask, Python, PythonAnywhere, REST API

• Shipped a production web app featuring 15 automated tools (merge, split, compress, image-to-PDF).

• Architected an ephemeral storage layer using Python/Flask, securely processing over 12,000+ documents with zero data retention via scheduled CRON cleanup.

• Scaled backend to comfortably serve 5,000+ organic users, optimizing PDF compression algorithms for 40% faster I/O.

View Live →

N8N AGENT

n8n, Telegram Bot, Gemini, Ollama, Gmail API

• Automated 20+ hours of weekly administrative tasks by architecting an asynchronous multi-turn AI assistant via n8n webhook pipelines.

• Implemented a sophisticated dual-LLM router (Gemini for high-throughput cloud tasks, local Ollama for zero-latency offline inference).

• Passed stateful conversation history seamlessly across a stateless webhook infrastructure.

View Repo →

BEAM

WebRTC, Flask-SocketIO, Render

• Engineered a low-latency P2P file transfer protocol utilizing direct WebRTC data streams, achieving stable transfer speeds up to 50MB/s.

• Overcame severe performance bottlenecks by implementing custom backpressure chunking algorithms, completely eliminating buffer overflow during multi-gigabyte transfers.

View Repo →

EXPERIENCE

AI/ML Engineer Intern

Edunet Foundation (TechSaksham - Microsoft & SAP)

Nov 2024 – Jan 2025

Tech Stack: Python, scikit-learn, Pandas, NumPy, Streamlit, NLP (CountVectorizer, Naive Bayes)

  • Spearheaded the development of an end-to-end NLP classification pipeline to detect sophisticated spam/phishing vectors, processing a raw dataset of 50,000+ localized emails.
  • Engineered rigorous feature extraction and hyperparameter tuning on a Multinomial Naive Bayes model, pushing the final F1-score to 96.4% and dropping false-positive rates by 22%.
  • Architected and deployed a containerized Streamlit microservice interface, reducing classification latency to <200ms and eliminating manual review overhead for non-technical operations staff.
  • Implemented automated Pickle-based model serialization strategies, establishing a version-controlled repository for model artifacts that accelerated iterative retraining workflows by 40%.

ABOUT & EDUCATION

Final year IT Engineering student in Pune, obsessed with the gap between a local demo and something people open at 2am. Makes engineering calls from measured results. Anime + solo bike rides.

Education & Certs

B.E. Information Technology
Sinhgad College of Engineering (SPPU) · 2022–2026

Certifications
Microsoft & SAP AI Transformative Learning
Godrej Infotech Data Analytics

SKILLS

AI/LLM

Ollama, Groq (Llama-3.3-70b), Gemini Flash, tool calling, fallback chain design, Tavily web search

BACKEND

Python, Flask, REST APIs, JWT auth, Flask-Limiter, CORS, Supabase postgres, SQLite, AWS EC2/S3

AUTOMATION & ML

n8n (webhook → AI agent pipelines), scikit-learn, CountVectorizer, Naive Bayes, pandas, NumPy, Telegram/Gmail Integrations

FRONTEND & TOOLS

HTML, CSS, JavaScript, PWA, Service Workers, IndexedDB, Git, Nginx, Gunicorn

CONTACT

YASH.GANGURDE001@GMAIL.COM LINKEDIN GITHUB TRY ARCANE →