01Projects

SWE, AI, and Data Science — built to solve real problems.

Six projects across full-stack engineering, security systems, computer vision, and voice AI. Each one built from a real constraint.

[ Sec. 02 ] Software Engineering

2024 — 2025 · Full-stack · Ride aggregation

RideHub

Unified ride aggregation platform combining five transport services — buses, cabs, metro, autos, and mixed routes — with real-road routing, booking history, and a FastAPI + PostgreSQL backend. Built to solve the transport unification problem in Indian cities.

Transport services unified

  • Next.js 15
  • FastAPI
  • PostgreSQL
  • React Native
View on GitHub

2025 · Security · Blockchain audit

RBAC-DocEncryption

Blockchain-audited document security system with AES-256-GCM field-level encryption, SHA-256 tamper detection, and role-based selective decryption. Designed for enterprise document workflows requiring immutable audit trails.

256-bit

Field-level encryption

  • Python
  • AES-256-GCM
  • SHA-256
  • Blockchain
View on GitHub

2025 · Healthcare · SIH 2025

NABHA Health

Telemedicine platform for rural India built for Smart India Hackathon 2025. Voice-based AI intake via Twilio + Bolna, adaptive wristband health monitoring, separate doctor and pharmacy interfaces, and offline-first design for low-connectivity areas.

173

Villages served (target)

  • React
  • FastAPI
  • Twilio
  • Bolna AI
  • Gemini API
View on GitHub

[ Sec. 03 ] AI & Data Science

2025 · Data Science · Forecasting

Walmart Sales Prediction

Stacking regressor pipeline combining XGBoost and Random Forest with feature engineering and EDA. Achieved R² of 0.965 and strong MAE reduction over a linear baseline across seasonal segments.

0.965

R² score

  • Python
  • XGBoost
  • Random Forest
  • Scikit-learn
  • Pandas
View on GitHub

2024 — 2025 · Computer Vision · CNN

Traffic Sign Recognition

CNN trained on 43 traffic-sign classes using augmented GTSRB data. Deployed low-latency inference with OpenCV for real-time video streams, with identical preprocessing enforced between training and inference to eliminate the dominant error source.

43

Sign classes

  • Python
  • TensorFlow
  • OpenCV
  • GTSRB
  • NumPy
View on GitHub

2025 · AI · Voice + RAG

NABHA Voice Intake

Voice-to-structured pipeline with retrieval-augmented summaries using Gemini API. Schema validation, confidence thresholds, and safe fallbacks ensure predictable UX. Error measured in user terms — time-to-correct and drop-off — not just model metrics.

RAG

Retrieval-augmented

  • Python
  • Gemini API
  • Twilio
  • RAG
  • FastAPI
View on GitHub