Building AI that
actually works.

Medical data is scattered and unstructured. LLMs hallucinate on medical facts without reliable grounding.

A RAG pipeline grounding LLM responses in verified medical documents — fast, accurate, hallucination-resistant.

1st place + ₹5,000 at SNS 8-hour hackathon. Iterated from a prototype built at Hack-A-Cure, VIT Chennai.

Journaling apps send data to the cloud. No existing voice tool runs offline and understands emotional context.

Offline-first Kotlin app that transcribes speech locally, detects emotional tone, and reflects it through the journaling experience.

IEEE ICECA 2025 published. Fully on-device NLP — zero data leaves the phone at any point.

Single-image SR loses spatial detail recoverable from paired satellite shots. Supervision data is rarely available at runtime.

Dual-image fusion deep learning model with attention architecture in PyTorch. No ground truth needed at runtime. Streamlit demo.

Handles real-world satellite pairs without paired supervision. Built end-to-end for Bharatiya Antariksh Hackathon.

Classifiers that retrain on every run are too slow for deployment. Preprocessing overhead adds unnecessary latency.

TF-IDF pipeline with optimized preprocessing that persists the trained model — inference runs without retraining.

Fast, production-ready inference designed with real deployment constraints in mind from the start.

Needed a real full-stack system covering seat selection, user accounts, and booking history.

Full-stack platform with user dashboard, booking flow, and history tracking. Frontend to backend on XAMPP.

Working system demonstrating full-stack capability across frontend, backend logic, and relational DB design.

No-code tools are desktop-only and complex. Building a site from a phone is still inaccessible for most people.

Flutter mobile app for drag-and-drop website creation with accessibility-first UI/UX design throughout.

Actively in development. Core layout engine and component system in progress.

Single-sensor flood monitors miss hidden risks and produce false alarms in remote deployments.

TTGO T-Call float combining water pressure, turbidity, and TDS sensor fusion into a Flood Risk Score, plus GSM alerting for warning and danger states.

Robust edge monitoring design for real-time flood awareness. Ideal for remote sites with low-connectivity GSM reporting and low-power deployment.

Traditional soil sensors leave hardware waste in fields. Retrieval at scale is costly and impractical for farmers.

Biodegradable sensor patches embedded into soil — monitor conditions in real-time, then decompose naturally. Zero retrieval, zero hardware waste.

Enables dense IoT deployment at agricultural scale without the environmental or logistical cost of hardware recovery.