Sarthak Jadhav

About

Artificial Intelligence & Machine Learning undergraduate with hands-on experience building agentic AI systems, LLM-integrated applications, and machine learning models from first principles. Actively building projects in public through hackathons and self-directed work while pursuing formal coursework in AI/ML.

Experience

• Technical Volunteer
  GDGoC WOW Pune
  Apr 2026 – 2026

Education

• B.Tech in Artificial Intelligence & Machine Learning
  PES Modern College of Engineering, Pune
  Artificial Intelligence & Machine Learning · Aug 2026 – Jul 2029

Skills

• SQL
• C
• Python
• Matplotlib
• NumPy
• Pandas
• Streamlit
• scikit-learn
• PyTorch
• TensorFlow
• Cli

Projects

• Student Performance Predictor with LLM Integration ↗
  Python, Streamlit, Pandas, LLM API

  Built an end-to-end pipeline that ingests student data, predicts performance, and generates personalised study recommendations via LLM API calls with engineered prompts. Deployed as a live Streamlit web app with real-time input/output for students and educators.

• Deep Learning from Scratch ↗
  Python, NumPy

  Implemented a fully connected neural network from first principles with forward propagation, backpropagation, and gradient descent, without ML framework dependencies. Engineered activation functions, weight initialization, and loss computation manually; benchmarked accuracy against Scikit-learn’s MLPClassifier across multiple datasets.

• TokenLean Agent — Context-Pruning Coding Agent ↗
  Python, Paritok API, rich/textual

  Designing and building an agentic coding loop (read, grep, edit, run-shell) with tool-calling, allow-listed tool executors, and a persistent context store for the Build with Paritok token-efficiency hackathon. Built a live CLI token-usage dashboard benchmarking pruned vs. naive full-history token consumption and routed accumulated context through a hosted GPU pruner before every LLM call.

Courses & certifications

• Machine Learning Specialization · Andrew Ng (DeepLearning.AI / Stanford Online, Coursera)

🏆 Achievements & awards

• Solved 50+ Data Structures & Algorithms problems on LeetCode

  Covered Arrays, Strings, Two Pointers, and Sliding Window.

• Self-taught prompt engineering and LLM API integration

  Built production-ready AI features outside the curriculum.

• Implemented backpropagation and gradient descent from first principles

  Built neural network training logic without relying on ML frameworks.