Projects

A snapshot of systems, libraries, and collaborations. Each project focuses on clarity, reproducibility, and measurable impact.



CaptionKraft

Built an end-to-end image captioning system that takes images as input and generates natural language descriptions using a CNN–LSTM model with attention. Deployed with Streamlit and MLOps tools (Terraform, Docker, Kubernetes) to enable scalable, cost-effective dataset generation and captioning workflows.

GitHub

Vital Vector

Developed a custom RAG-based fitness & nutrition assistant that provides personalized health guidance using a local LLM, semantic search over expert textbooks, and Gradio UI—tailored to user profiles with upcoming tool-calling for real-time videos fetching support.

GitHub Live Demo

Jarvis

Built a multimodal WhatsApp-based AI assistant with voice, image, and text understanding using LLaMA‑3, LangGraph, Whisper, and Qdrant—deployed via Docker & GCP with persistent memory, real-time responses, and CI/CD automation for scalable intelligent interaction.

GitHub

Image Forensics

Developed a CNN-based system to detect real, AI-generated, and edited images with 96% accuracy; enhanced detection by training on synthetic images from DCGANs and ensured robustness through extensive unit testing.

GitHub

UberFlow Analytics

Built a complete data analytics pipeline using Mage, BigQuery, and Looker to ETL and visualize NYC TLC trip data; delivered an interactive dashboard with dynamic charts, maps, and filters to support operational insights and decision-making.

GitHub

FormFix

Developed a real-time exercise form correction tool using OpenCV, Mediapipe, and CNNs to track body landmarks and provide instant feedback—enhancing safety and reducing injury risk through posture-aware guidance.

GitHub