Python · Scikit-learn · NLP · Streamlit

Engineered a disease-prediction chatbot using Random Forest with confidence scoring. Built an NLP pipeline with synonym mapping & fuzzy matching for accurate symptom extraction.

Python · Pandas · Scikit-learn · Matplotlib

Benchmarked Logistic Regression & Random Forest models achieving a ROC-AUC of 0.90. Handled end-to-end preprocessing — missing values, encoding, and feature scaling.

Python · Flask · Scikit-learn · Google Maps API

Built a real-time demand-forecasting system analyzing traffic, geolocation & battery data. Integrated Google Maps API for station visualization & route optimization.

Python · LangChain · Qdrant · LLM APIs

Built a Retrieval-Augmented Generation system that chunks & embeds documents into a vector store, then answers natural-language queries with grounded, citation-backed responses — cutting hallucination on domain-specific questions.

Python · YOLO · OpenCV · ONNX

Deployed a real-time object detection pipeline using YOLO, optimized with ONNX runtime for low-latency inference on live video streams — used for tracking and counting objects in a video feed.

Python · Embeddings · Vector Search · 🏆 1st — Anveshan 3.0, IIIT Delhi

An AI-powered "second brain" that auto-links related notes using semantic embeddings, generates concept maps, and surfaces spaced-repetition prompts to reinforce long-term recall.

Python · Behavioral ML · Real-Time Tracking · 🏆 1st — Hack Sphere 2024, ARSD College

A productivity companion that detects distraction patterns from real-time activity signals and nudges users with adaptive break/focus cycles to maximize deep-work sessions.

Python · NLP · PDF Parsing · 🥈 2nd — HP Global AI for Education 2024

An AI tool that ingests course material and auto-generates topic-aligned quizzes with difficulty calibration, answer keys, and explanations — cutting manual quiz creation time for teachers.

GPT-4 · Stable Diffusion · Prompt Engineering · 🏅 4th — GPT-4 x SD Hackathon 2023

A unified pipeline where GPT-4 expands rough prompts into rich, detailed image prompts, feeding into Stable Diffusion to generate refined, styled visuals from a single idea.

Python · Time-Series Forecasting · Sensor Telemetry · 🏅 7th — Ametcs Global GenAI 2024

A predictive maintenance system analyzing vibration, temperature & usage-cycle telemetry to forecast component failure windows, enabling condition-based maintenance scheduling.

Python · NLP · Clause Extraction · 🥉 3rd — Pravaah AI Hackathon 2023

An AI assistant that parses legal documents, summarizes key clauses in plain language, and flags potentially risky or unusual terms for review — making contracts accessible to non-lawyers.

Python · Backtesting Engine · Strategy Orchestration · 🎖️ 56th — ETHIndia 2023

A platform that lets users combine multiple algorithmic trading/decision strategies into a single executable "combo," with backtesting against historical data before deployment.

🎯 Problem Students and professionals struggle to retain and connect knowledge scattered across notes, articles, and study material — leading to fragmented learning and poor long-term recall.

💡 Solution An AI-powered "second brain" that auto-links related notes using semantic embeddings, generates concept maps, and surfaces spaced-repetition review prompts to reinforce memory.

🛠️ My Contribution Designed and built the semantic similarity engine (embeddings + vector search) powering auto-linking, plus the backend APIs connecting the ML pipeline to the frontend.

🎯 Problem Remote learners and WFH professionals lose productive hours to digital distractions, with no real-time way to measure or correct focus loss during work sessions.

💡 Solution A productivity companion that tracks activity/behavioral signals to detect distraction patterns in real time and nudges users with adaptive break/focus cycles.

🛠️ My Contribution Built the ML component that classified user activity into focus/distraction states, and tuned the nudge-timing logic based on session data.

🎯 Problem Teachers spend hours manually creating quizzes and assessments aligned to specific chapters or learning outcomes, slowing down curriculum delivery.

💡 Solution An AI tool that ingests course material (text/PDFs) and auto-generates topic-aligned quizzes with varying difficulty levels, complete with answer keys and explanations.

🛠️ My Contribution Built the question-generation pipeline using NLP for content parsing and difficulty calibration, and integrated the PDF-to-quiz ingestion flow.

🎯 Problem Reading and understanding lengthy legal documents and contracts is time-consuming and error-prone for non-lawyers, leading to missed clauses and risks.

💡 Solution An AI assistant that parses legal documents, summarizes key clauses in plain language, and flags potentially risky or unusual terms for review.

🛠️ My Contribution Developed the document parsing and clause-extraction pipeline using NLP, and built the risk-flagging logic for non-standard clauses.

🎯 Problem Creators need a fast way to go from a text idea to a polished visual concept, without juggling multiple disconnected AI tools for prompting and image generation.

💡 Solution A unified pipeline where GPT-4 expands rough prompts into rich, detailed image prompts, feeding into Stable Diffusion to generate refined, styled visuals end-to-end.

🛠️ My Contribution Implemented the GPT-4 prompt-refinement layer and the integration logic connecting it to the Stable Diffusion generation pipeline.

🎯 Problem Unplanned aircraft component failures cause costly downtime and safety risks, and traditional scheduled maintenance often replaces parts too early or too late.

💡 Solution A predictive maintenance system analyzing sensor telemetry (vibration, temperature, usage cycles) to forecast failure windows and enable condition-based maintenance.

🛠️ My Contribution Worked on the time-series forecasting model for failure prediction and engineered features from raw sensor telemetry data.

🎯 Problem Combining multiple trading/decision algorithms into a single coherent strategy is complex — most tools don't let users blend or backtest algorithmic approaches easily.

💡 Solution A platform that lets users combine multiple algorithmic strategies into a single executable "combo," with backtesting against historical data before deployment.

🛠️ My Contribution Worked on the backtesting engine and helped implement the logic for combining multiple strategy outputs into a single decision signal.