Atharva Samant

💻 Projects

Things I've Built

Full-stack applications, AI/ML systems, and production-ready architectures.

🧠

MU-Cortex

AI Academic Intelligence Platform

2024 — Present 🔗 Live Site

Designed and deployed a full-stack AI platform for Mumbai University students
Built end-to-end architecture integrating automated PDF ingestion, LLM analysis pipeline, and behavioral analytics dashboard
Engineered FastAPI-based pipeline to extract, cluster, and map PYQ questions to syllabus subtopics using Gemini
Implemented semantic clustering, cross-year frequency analysis, and AI-powered answer generation
Developed Streamlit-based review tools for manual quality control before serving content
Designed analytics system using Supabase to track user behavior, feature adoption, and drop-offs
Implemented Row-Level Security (RLS), CORS handling, and secure API key management for production

Next.js FastAPI Gemini 2.5 Flash Supabase Streamlit

⚖️

AstraLex

AI Legal Assistant (RAG System)

2025 🐙 GitHub

Built a full-stack AI legal assistant for Indian law using an advanced Retrieval-Augmented Generation (RAG) pipeline
Designed adaptive pipeline to classify queries into fast, standard, and deep-research paths
Implemented hybrid retrieval (dense vector + BM25 + reranking) with Qdrant as vector database
Developed stateful chat system with persistent sessions, streaming responses, and context-aware reasoning
Built modular backend (FastAPI) handling auth, chat orchestration, document ingestion, and analytics
Designed PostgreSQL schema for users, sessions, messages, feedback, and logs
Implemented JWT authentication and query analytics pipeline for system monitoring

Next.js FastAPI Qdrant PostgreSQL JWT BM25 LLMs

⚙️

AutoML Platform

End-to-End ML Training & Deployment System

2025 🐙 GitHub

Built an end-to-end AutoML platform enabling dataset upload, asynchronous model training, and real-time inference
Designed training engine with Optuna-based hyperparameter optimization, multi-objective scoring, and Pareto-optimal model selection
Implemented weighted ensemble learning with SLSQP optimization and model retraining pipeline
Developed job orchestration system with background workers and real-time job tracking APIs
Built model registry system storing trained models, metadata, and feature importance for reproducible inference
Created full-stack interface (Next.js + Streamlit) for training visualization, leaderboard tracking, and predictions

FastAPI Optuna scikit-learn MLflow Next.js React Joblib

💍

AI Pre-Wedding Shoot Generator

Generative AI for Personalized Video Scenes

Aug 2025 — Present

Building a generative AI system to create personalized pre-wedding video scenes from user-uploaded images
Integrated Stable Diffusion XL with LoRA fine-tuning to preserve identity during generation
Implementing motion transfer and automated storyboard generation for coherent scene sequencing
Achieved high identity consistency with strong ClipFace similarity scores in early results

Python PyTorch SDXL LoRA FastAPI CUDA

🎓 Education

My educational background and accomplishments.

B.E. in Computer Science and Engineering (AI & ML)

A.P. Shah Institute of Technology, Mumbai

📅 2023 — Present GPA: 9.4 / 10.0 📍 India

Higher Secondary Certificate (HSC)

New English Jr. College

📅 2021 — 2023 Percentage: 79% 📍 Thane, India

Secondary School Certificate (SSC)

Carmel Convent High School

📅 2006 — 2021 Percentage: 86.40% 📍 Thane, India

🏆 Achievements

🏅

2024 — Participated in 2+ inter-college hackathons; built an AI-based forgery detection system within 24 hours.