Skip to content
View AviMath2412's full-sized avatar
🌴
On vacation
🌴
On vacation

Highlights

  • Pro

Block or report AviMath2412

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.

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
AviMath2412/README.md

> initializing avi.config.json ...
> Domain: AI × RL × Systems × Agentic AI
> Status: Building things that learn, adapt, and scale.

whoami

I'm Avi Mathur — an AI engineer obsessed with making machines reason, adapt, and act.

  • 🏆 Apple Swift Student Challenge 2026 Winner — built ScanSafe, an on-device allergen scanner powered by Apple Intelligence & VisionKit
  • 🤖 I implement RL algorithms from scratch — Bellman equations, policy gradients, the whole math
  • 🌱 Currently designing RL training environments for knowledge-work agents
  • 🔬 Research interests: Reinforcement Learning · Agentic Systems · NLP · AI for Social Good
  • 📬 Reach me: mathuravi668@gmail.com


🧠 Research Projects

📨 Email Triage RL Environment 2026

OpenEnv · Python · FastAPI · Docker · HF Spaces

Modelled email triage as a high-stakes RL task with 3 difficulty tiers. Designed a dense step-wise reward function decomposed into correctness, efficiency & routing accuracy — tackling the sparse-reward bottleneck in knowledge-work agents.

GPT-4o-mini zero-shot results:

  • Easy: ~94% pass rate
  • Medium: ~80% pass rate
  • Hard: ~90% pass rate

🎮 DDPG — From Scratch 2025

Python · TensorFlow · RL Theory

Implemented Deep Deterministic Policy Gradient from first principles — actor-critic updates, experience replay, soft target network updates derived directly from the original paper.

No library abstractions. Pure math.

Bellman equations · Policy gradient theorem · Off-policy stability

🧬 NeuroScan AI 2025

Python · TensorFlow · Docker · CNN

High-accuracy CNN for medical image classification on imbalanced clinical data. Studied architectural tradeoffs: depth, skip connections, pooling. Containerized with a real-time prediction interface.

🌾 Multilingual Mandi 2024

JavaScript · Ollama (LLaMA 3.2) · Multilingual NLP

Farmer-first AI platform delivering real-time mandi price queries in Indian regional languages. Voice-based queries with on-device NLP via Ollama — built to work in low-connectivity environments.


⚙️ Tech Stack

AI / ML / RL

Python TensorFlow NumPy Pandas Scikit-learn

Systems & Deployment

Docker FastAPI Hugging Face GCP Oracle Cloud

Languages & Web

C++ TypeScript Next.js React PostgreSQL


🏆 Achievements

Apple Swift Student Challenge 2026 Winner Selected globally for ScanSafe — on-device allergen scanner using Apple Intelligence, VisionKit & Spatial UI
🤖 Meta PyTorch OpenEnv Hackathon x Scaler Cleared Round 1 out of 52,000+ developers — headed to the Grand Finale in Bangalore, sponsored by Meta, PyTorch, and Hugging Face
🏦 SBI Life Hack-AI-Thon Led technical team as a national hackathon finalist
💼 EY Techathon 5.0 Technical team lead
🇮🇳 Smart India Hackathon 2025 Volunteer, Grand Finale

📜 Certifications

  • 🟠 Oracle Cloud Infrastructure 2025 — Certified Generative AI Professional
  • 🟠 Oracle AI Vector Search — Certified Professional
  • 🔵 Walmart USA Advanced Software Engineering — Virtual Experience (Jan 2026)

🌐 Connect

LinkedIn GitHub Portfolio Email X YouTube


"I don't just use ML — I understand it from the math up."

Pinned Loading

  1. Multilingual-Mandi Multilingual-Mandi Public

    Multilingual Mandi is a farmer-first web platform that delivers AI-powered real-time mandi prices, voice-based queries, and smart negotiation in multiple Indian languages - designed to make local t…

    JavaScript 1

  2. Myportfolio Myportfolio Public

    TypeScript

  3. NeuroScan-AI NeuroScan-AI Public

    Python 1