Diwanshu Gangwar

Software Engineer, AI Engineer, Machine Learning Engineer

80 · Excellent

About

B.Tech graduate in Artificial Intelligence and Data Science with experience building full-stack applications, machine learning systems, and AI-powered software. Worked with Python, TypeScript, SQL, Next.js, Node.js, TensorFlow, and modern AI frameworks to develop real-time applications, REST APIs, retrieval-augmented systems, and deep learning models. Interested in Software Engineering, AI Engineering, and Machine Learning roles.

Skills & Expertise (37)

Python Expert

9.0/10

Years Exp

Deep Learning Expert

9.0/10

Years Exp

NLP Advanced

8.5/10

Years Exp

React Advanced

8.5/10

Years Exp

TensorFlow Advanced

8.5/10

Years Exp

Node.js Advanced

8.0/10

Years Exp

FastAPI Advanced

8.0/10

Years Exp

JavaScript Advanced

8.0/10

Years Exp

SQL Intermediate

7.5/10

Years Exp

Git Intermediate

7.5/10

Years Exp

Docker Intermediate

7.0/10

Years Exp

Data Structures & Algorithms Object-Oriented Programming DBMS VS Code Operating systems Postman Jupyter Notebook Streamlit LINUX GitHub Pydantic API Integration REST APIs LLM Applications Retrieval-Augmented Generation Transformers Transfer Learning FAISS SQLite Postgresql Socket.IO Next.js LangChain scikit-learn Typescript C++

Work Experience

Developer

BlinkTalk

Present - Present

Architected a full-stack anonymous video chat platform using Next.js 16, Node.js, and Socket.IO, enabling real-time browser-based peer matching with zero sign-up friction. Engineered a server-side queue-based matchmaking system that handles 1-to-1 room creation, automatic partner reassignment, and live session teardown via in-memory pair tracking. Integrated ZegoCloud WebRTC SDK for low-latency video rooms with lazy initialization, reducing unnecessary SDK load until a peer match is confirmed. Built a fully responsive, component-driven UI using TypeScript, Tailwind CSS 4, and Framer Motion, improving interaction clarity across desktop and mobile viewports.

Developer

AI Resume Analyzer

Present - Present

Built an end-to-end Retrieval-Augmented Generation (RAG) pipeline using LangChain, FAISS, and Google Gemini API to perform semantic comparison of resumes against job descriptions. Implemented a 4-stage document intelligence pipeline — PDF parsing, text chunking (500-char chunks, 50-char overlap), vector embedding, and FAISS-indexed semantic retrieval — reducing manual screening effort. Automated model discovery logic to select the best available Gemini version (2.0-flash → 1.5-flash → 1.5-pro), with keyword-based heuristic fallback ensuring 100% uptime even when API is unavailable. Deployed a Streamlit web interface supporting resume upload, targeted Q&A, improvement suggestions, and structured skill-gap report generation for recruiters and job seekers.

Developer

Chest X-ray Disease Classifier

Present - Present

Trained a DenseNet121 transfer learning model on 9,298 labeled chest X-ray images (Normal / Pneumonia / Tuberculosis), achieving 91.78% test accuracy and 0.992 weighted AUC on a held-out test set. Applied class-weighted training, image augmentation, and fine-tuning on 2.16M trainable parameters to address class imbalance and improve generalization across underrepresented disease categories. Attained per-class ROC-AUC scores of 0.988 (Normal), 0.990 (Pneumonia), and 0.998 (Tuberculosis), demonstrating high sensitivity for rare but critical diagnoses. Deployed the model as a Streamlit application with an average inference latency of ∼403 ms/image on CPU, enabling real-time diagnostic support without GPU infrastructure.