# Harshdip Saha

> Harshdip Saha — machine-learning researcher in medical brain imaging (NexGenLab NSUT, IIT Madras). RECAP-Net placed 3rd worldwide in the BraTS Lighthouse 2025 challenge and was presented at MICCAI 2025. Open to SDE and research internships.

Harshdip Saha — UG researcher, NexGenLab NSUT, New Delhi, India. This file is the machine-readable index of https://harshdipsaha.tech. It is generated at build time from the same content the site renders, so it cannot drift from the published pages.

## Site

- [Home](https://harshdipsaha.tech/): Overview: the research, the work, and how it is made.
- [Story](https://harshdipsaha.tech/story): I'm a pre-final-year Computer Science student, specialising in AI, at Netaji Subhas University of Technology in Delhi, and an undergraduate researcher at NexGenLab NSUT. My work sits where machine learning meets medicine: computer vision, brain imaging, and the pipelines that make a model's answer trustworthy enough to check against a radiologist's.
- [Projects](https://harshdipsaha.tech/projects): Every project, newest first, each with a written case study.
- [Gallery](https://harshdipsaha.tech/gallery): Photographs.
- [Process](https://harshdipsaha.tech/process): How every change to this site is planned, recorded and gated (AI-DLC).
- [Ask AI](https://harshdipsaha.tech/ask-ai)

## Projects

- [BrainwavesFinland — finding a tumour with microwaves](https://harshdipsaha.tech/projects/brainwavesfinland): Microwave tomography: locating a 30 mm tumour inside a brain phantom from antenna measurements alone. Reconstructed to 1.4 cm median error and 2° angular error across ten positions. (2026)
- [PyAMorph (formerly pySdf)](https://harshdipsaha.tech/projects/pysdf): High-fidelity implicit geometry modeling: 2D/3D signed distance functions, microstructure generation, and FFT homogenization in Python. (2026)
- [SAAKSHI — the witness for every rupee](https://harshdipsaha.tech/projects/saakshi): A read-only forensic layer over India's public spending data, plus a toll-free voice line so any citizen can ask what was spent in their village. Hackathon prototype for TechGig × Optum's Inclusive Innovation for Bharat. (2026)
- [tinySafetyNet](https://harshdipsaha.tech/projects/tinysafetynet): No internet and feeling unsafe? Keep this tiny thing in your pocket. (2026)
- [Zombies-learning-progression-PGGAN](https://harshdipsaha.tech/projects/zombies-learning-progression-pggan): Creating zombies similar to humans — PGGAN for temporal brain MRI imaging. (2026)
- [BrandDiffusion](https://harshdipsaha.tech/projects/branddiffusion): Context-aware, brand-safe creative automation — winner, Adobe Express Add-ons for Enterprise Hackathon. (2026)
- [Agentic Loan Assistant Chatbot](https://harshdipsaha.tech/projects/agentic-loan-assistant-chatbot): EY Hackathon 2025 — Agentic loan assistant chatbot. (2025)
- [AtoM-Net: Cognitive Neuro-Symbolic Pipeline for Negotiation Theory of Mind](https://harshdipsaha.tech/projects/atomnet): Neuro-symbolic pipeline combining LLM belief inference with a symbolic decision engine for negotiation theory of mind. In review at EMNLP 2026. (2025)
- [Aquila Optimiser Optimised](https://harshdipsaha.tech/projects/aquilaoptimiseroptimised): Can Aquila catch prey with more precision? Evolutionary optimization. (2025)
- [APT — Accurate Precise Timely](https://harshdipsaha.tech/projects/apt): HealthRevolution — AI-powered healthcare scheduling. Top 30 at AI4Humanity Summit. (2025)
- [Tomato Disease Detection](https://harshdipsaha.tech/projects/tomato-disease-detection): CNN classifier for tomato leaf diseases, deployed as a Streamlit app. (2025)
- [Object Tracking — Tennis Game](https://harshdipsaha.tech/projects/object-tracking-tennis-game): Player and ball tracking with real-time speed and mini-court stats overlay. (2025)
- [Online Museum Ticketing Chatbot](https://harshdipsaha.tech/projects/online-museum-ticketing-chatbot): Avoid crowds — use our AI bot for museum ticketing. (2025)
- [AI-Enhanced Healthcare Diagnostics](https://harshdipsaha.tech/projects/ai-enhanced-healthcare): AI-driven healthcare diagnostics and management system. (2025)
- [AI-Generated Text Detector](https://harshdipsaha.tech/projects/ai-generated-text-detector): Fine-tuned models for AI-generated text detection. (2024)
- [Missing Person Identification in Crowds](https://harshdipsaha.tech/projects/missing-person-identification): AI-powered missing person detection in crowds. Top 45 at Smart India Hackathon 2024. (2024)
- [Anime Recommender](https://harshdipsaha.tech/projects/anime-recommender): Collaborative-filtering anime recommendations, ranked by cosine similarity over the MyAnimeList dataset. (2024)
- [GUI CANSAT](https://harshdipsaha.tech/projects/gui-cansat): Tkinter GUI that replays CanSat flight telemetry as live plots. (2024)
- [Alien Invasion (Pygame)](https://harshdipsaha.tech/projects/alien-invasion-pygame): Space-Invaders-style arcade shooter — first project with Pygame. (2024)
- [YouTube Comment RAG](https://harshdipsaha.tech/projects/youtubeproj-langchain): Consensus-weighted RAG that answers what a comment section actually thinks, not five cherry-picked comments. (2024)

## Research

- [RECAP-Net: does the tumour grow? Classifying glioblastoma response from paired MRI scans](https://link.springer.com/10.1007/978-3-032-16370-7_23): MICCAI 2025, South Korea — oral presentation. Published in Springer Lecture Notes in Computer Science. 3rd worldwide, BraTS Lighthouse 2025 Tumor Progression Challenge — every team's model scored on the same hidden brain-tumour scans.
- [RECAP-Net: does the tumour grow? Classifying glioblastoma response from paired MRI scans — Code](https://github.com/HARSHDIPSAHA/brats_response_project)
- [Top open-source contributor, BrainGlobe](https://github.com/brainglobe/brainglobe-registration): 15+ merged pull requests across the ecosystem; leading contributor to brainglobe-registration, the tool that aligns a lab's brain images to a reference atlas.
- [All India Rank 14 — BRAINDEAD, Unstop](https://github.com/HARSHDIPSAHA/braindead_1-2): A national data-science competition. Two builds in one entry: a data-analysis-and-prediction pipeline and a text summariser that pairs a language model with classical methods.
- [Top 30 of 150+ teams — AI4Humanity Summit](https://github.com/HARSHDIPSAHA/APT): Pitched Accurate Precise Timely, an AI-powered hospital scheduling system, in person at the summit held with the Israel Embassy.

## Elsewhere

- [GitHub](https://github.com/HARSHDIPSAHA): Source for nearly every project listed above.
- [LinkedIn](https://www.linkedin.com/in/harshdip-saha): Professional profile.
- [Résumé (PDF)](https://harshdipsaha.tech/resume.pdf): One-page CV.
- [Email](mailto:harshdipsaha@gmail.com): Direct contact — harshdipsaha@gmail.com.

## Optional

- [llms-full.txt](https://harshdipsaha.tech/llms-full.txt): This same index with the full text of every project case study inlined — one request for the whole site.