Anshuman Singh

My Experience

Internship

AI-ML Virtual Intern

EdiSkills Foundation

Jan 2026 - Mar 2026

Participated in a 2-month, industry-aligned virtual internship supported by Google for Developers and AICTE, focusing on end-to-end Machine Learning lifecycles.
Engineered on-device AI applications using ML Kit, implementing high-performance features such as Real-time Face Detection, Text Recognition (OCR), and Object Tracking.
Integrated ML Kit's Vision APIs to process live camera streams, enabling real-time interaction between user physical inputs and AI-powered game logic.

Project Lead

Student ERP system at Haridwar University

Backend Developer

2026 - Present

Architected and implemented the complete backend database schema and end-to-end system workflow.
Enhanced system performance and reduced query latency by implementing caching mechanisms and strategic database indexing.
Developed an AI-powered engine for automated routine generation to optimize academic scheduling and administrative tasks.

Featured Projects

HealthTech

DocAI

A medical intelligence platform built with React (Vite) and Flask, featuring a dynamic ReAct Agent architecture. DocAI leverages MedGemma-4b for agentic reasoning and Qwen2.5-7B for background clinical analysis, backed by PostgreSQL + pgvector for precise BioBERT embedding retrieval.

React (Vite) PostgreSQL + pgvector BioBERT (768D) MedGemma-4b-it Qwen2.5-7B ReAct Agent

GitHub Code Live Demo

AI & Optimization

AI Timetable Generator

Automated scheduling system using Google OR-Tools (CP-SAT solver) to generate optimal institutional timetables while adhering to complex hard and soft constraints.

Constraint Programming Universal CSV Ingestor Intelligent Bottleneck Detection Google OR-Tools

GitHub Code Live Demo

Computer Vision

Traffic Sign Recognition

A deep learning system for high-accuracy recognition and classification of 43 different traffic sign categories. Built with a custom Convolutional Neural Network (CNN) architecture using TensorFlow and OpenCV, trained on the GTSRB dataset.

TensorFlow/Keras OpenCV CNN Architecture GTSRB Dataset

GitHub Code

Smart City / AI

CBMS

An advanced hybrid-architecture computer vision pipeline designed to track individuals and detect civic offences in near real-time. Features dual dashboards (Admin & User Hub) with automated evidence generation and 3D skeletal behavior classification.

YOLOv8 + ByteTrack InsightFace (Face ID) MediaPipe + Bi-GRU Hybrid Cloud-Edge

GitHub Code

Edge ML / Optimization

Edge AI Optimization

A complete pipeline for compressing deep learning models using INT8 Post-Training Quantization. Reduces MobileNetV2 storage footprint by ~70% while maintaining inference accuracy for deployment on memory-constrained IoT and mobile devices.

TensorFlow Lite Post-Training Quantization INT8 Weight Mapping MobileNetV2

GitHub Code

Model Compression

Mobile / Computer Vision

Product Image Search

An Android application leveraging Google's ML Kit to identify fashion products in real-time. Features interactive object selection, dynamic image processing, and integration with Vision Product Search for instant visual results.

Kotlin Google ML Kit Object Detection Vision Search API

GitHub Code