I'm a Final Year Computer Engineering student at Fr.Conceicao Rodrigues College of Engineering, Mumbai with a CGPA of 9.15, passionate about building systems that sit at the intersection of research and real-world deployment. I design end-to-end intelligent applications that unify robust full-stack architecture with cutting-edge GenAI — spanning TinyML, Computer Vision, NLP, and Explainable AI (XAI).

Backed by peer-reviewed conference research and a published patent, I build:

• 🔬 Lightweight deep learning architectures optimized for edge deployment
• 🌐 Full-stack MERN applications with clean, scalable system design
• 🛠️ Local-first developer tooling that prioritizes privacy and performance
• ⚙️ High-performance systems in C++ with compiler-level optimizations

📍 Location    : Mumbai, Maharashtra, India
🎓 Degree      : B.E. Computer Engineering — FRCRCE Mumbai (Expected 2027)
📊 CGPA        : 9.15 / 10.0
🔭 Focus Areas : AI/ML Research · Full-Stack Engineering
🚀 Currently   : Publications in progress after presenting at ICASET & ICRETM 2026

ICASET 2026  |  Presented · Publication In Progress

TinyML XAI EEG Temporal Attention

A lightweight deep learning model for single-channel EEG sleep staging that replaces recurrent networks with a temporal attention mechanism — enabling deterministic model interpretability and XAI visualization.

• Engineered a frequency-sensitive feature extractor for edge-constrained inference
• Benchmarked on the Sleep-EDF dataset with competitive accuracy at minimal compute
• Purpose-built for TinyML deployment

ICRETM 2026  |  Presented · Publication In Progress

EfficientNet-B0 ArcFace Transfer Learning Edge AI

Fine-grained classification for 16 native Indian bovine breeds using EfficientNet-B0 + ArcFace (Additive Angular Margin Loss).

• +39% F1-score improvement for visually similar breeds (Mehsana buffalo)
• 54MB model footprint · 50ms inference latency — offline mobile ready
• Deployed for Bharat Pashudhan mission: fair livestock pricing & insurance verification

TypeScript Python torch.fx MCP React

A VS Code extension for ML architecture visualization using torch.fx symbolic tracing and a custom AST-based Topological Shape Propagator for deterministic graph reconstruction.

• Model Context Protocol (MCP) — local-first, zero-egress, full data privacy
• Glassmorphism webview for dynamic tensor retracing & real-time dimension editing
• Eliminates external cloud inference during architecture exploration
• Automatically converts complex PyTorch AI code into interactive visual graphs
• Helping developers easily debug and trace machine learning models in real time

Python DistilBERT Hugging Face NLP

Fine-tuned and open-sourced a DistilBERT model for mental health NLP classification.

• 97.13% accuracy & F1-score across 50,000 balanced samples
• Binary risk-detection pipeline: Normal / Risk Detected classification
• Optimized for high-precision screening in social media & clinical NLP pipelines

🔒 Private Repository — IP Protected

C Flex Bison Compiler Design Node.js

A proprietary JSON query engine with a full compiler pipeline: lexical analysis → syntactic parsing → AST construction → AST-level Filter Promotion optimization.

• Sub-200ms query latency on nested JSON workloads
• Interactive web UI tracing raw query → AST transformation
• Built to demystify compiler internals for systems programming education

React FastAPI MongoDB JWT

A full-stack academic resource allocation platform on the MERN-adjacent stack.

• A full-stack university app (React/FastAPI) to automate course allocation. Students submit preferences and admins run the allocation algorithm.
• Stateless JWT authentication with role-based access control
• 🔗 View on GitHub

🏠 Project K.A (Kitchen Automate)  —  IoT & AI-Based Smart Kitchen Trolley

Published Patent  |  IoT AI

The Smart Kitchen Inventory and Recipe Suggestion System is an IoT-enabled kitchen management solution designed to simplify meal planning and inventory tracking. Equipped with weight sensors, cameras, and a mobile app interface, this smart system automates ingredient monitoring, offers personalized recipe recommendations based on available stock, and generates grocery lists to streamline shopping. The system includes a sensor-enabled kitchen trolley with dedicated compartments, a cloud-based backend for data processing, and a mobile app for real-time updates, low-stock alerts, and meal suggestions. By combining sensor data with image recognition, the system ensures precise inventory management, minimizes food waste, and promotes ingredient freshness — creating a more efficient and enjoyable cooking experience.

Languages

AI / ML & Research

Hugging Face Transformers  ·  TinyML  ·  Computer Vision  ·  NLP  ·  XAI

ArcFace  ·  torch.fx  ·  DistilBERT  ·  EfficientNet  ·  Transfer Learning

Streamlit  ·  Model Context Protocol (MCP)

Full-Stack & MERN

Systems & Compiler Engineering

Flex / Bison  ·  AST Design  ·  Compiler Optimization  ·  Systems Programming

Cloud, DevOps & Tools

 

💡 Open to research collaborations, SDE & ML Engineering roles, and high-impact engineering conversations.