Welcome to my portfolio

Chandan Kumar

AI · ML · Data Enthusiast

ML Engineer building end-to-end model training and data pipelines across deep learning architectures (GNN, LSTM, UNet, transformers) and LLM/RAG systems. Hands-on with multilingual NLP (Indic-language RAG), open-source tooling (published PyPI package), local LLM serving (vLLM, ONNX Runtime), and large-scale experimentation across 30+Indian cities. Comfortable taking models from prototype to production.

About Me

Technical Expertise

Specialized in building intelligent systems using cutting-edge AI/ML technologies. Proficient in deep learning frameworks, LLM applications, and MLOps practices.

Innovation Focus

Passionate about implementing advanced RAG techniques, building agentic AI systems, and creating data-driven solutions that solve real-world problems.

Work Experience

ML Engineer
Indian Institute of Technology Delhi
May 2026 - current
Developing hydrological models using hybrid Graph Neural Network (GNN) and Long Short-Term Memory (LSTM) architectures.
AI Engineer
AI Center of Excellence - IIT Gandhinagar
Nov 2025 - April 2026
Worked on data preprocessing pipelines and scenario-based flood modeling across different conditions
Built and experimented with models like U-Net and GNN, incorporating physics-informed constraints for improved realism and generalization
Curated datasets independently by transforming numerical data into structured textual formats, integrating multiple sources
Integrated AI-driven analytics into a multi-layer geospatial dashboard, enabling natural language querying over structured spatial data
Sr. Project Technical Assistant
Indian Institute of Technology Bombay
Sept 2024 - Oct 2025
Developed real-time data pipelines in Python for streaming sensor data and integrated predictive time-series models on AWS EC2
Built automated pipelines for large-scale satellite-based atmospheric data downloading, preprocessing, and integration
Developed a data cleaning and standardization module for CPCB ground station data, released as open-source package on PyPI (airpy-tool)
Designed comprehensive ML pipeline with hyperparameter tuning (GridSearchCV, Optuna, Grey Wolf Optimization) for PM2.5 forecasting across 30 Indian cities
Data Science Intern
Jal Jeevan Mission, IIM Bangalore
Jan 2024 - Aug 2024
Automated data scraping tasks using Selenium and BeautifulSoup, significantly improving efficiency and accuracy of data collection
Built and deployed district-level dashboards for Karnataka on organization's Windows Server
Developed an Indic languages AI chatbot system using RAG architecture, trained on domain-specific and customized data

Technical Skills

Programming

Python Object-Oriented Programming Data Structures and Algorithms

Machine Learning & Deep Learning

Scikit-learn TensorFlow PyTorch Keras Ray Tune Optuna LSTM GNN

LLMs & Agentic AI

LangChain LangGraph LlamaIndex Vector Database RAG Pipeline Prompt Engineering MCP

MLOps & Deployment

FastAPI Flask Docker MLflow Weights & Biases Git CI/CD with GitHub Actions ONNX Runtime Kubernetes (basics)

Data Processing

NumPy Pandas PySpark GeoPandas Folium Web Scraping

Cloud & Databases

AWS (EC2, S3, Lambda) MongoDB SQLite NoSQL Satellite & Earth Observation Data Pipelines

Featured Projects

Advanced RAG

A sophisticated PDF question-answering chatbot that implements multiple advanced Retrieval-Augmented Generation (RAG) techniques for improved accuracy and reasoning capabilities.

LangChain LlamaIndex Vector Database Knowledge Graph Chain-of-Thought

Key Highlights:

  • Designed modular RAG system adapting to query complexity using multiple reasoning strategies
  • Implemented Auto Mode to dynamically select appropriate retrieval and reasoning strategy
  • Developed session-scoped memory to maintain conversational context and improve multi-step query handling

Advanced RAG Techniques:

LLM-Powered Decomposition

Breaks complex questions into simpler sub-questions

Chain-of-Thought (CoT) Prompting

Uses step-by-step reasoning to answer questions

Graph-Based Reasoning

Builds and queries a knowledge graph of entities and relationships

Auto Technique Selection

Automatically chooses the best technique based on question complexity

Route-Planner

Built a train search app with AI-powered personalized recommendations for travel planning.

Playwright vLLM FastAPI Local LLM

Key Highlights:

  • Built train search using Playwright for faster scraping, cutting search time by 3-4x compared to Selenium
  • Served local LLM via vLLM for inference, generating personalized train recommendations based on user preferences
  • Exposed search and recommendation features through a versioned FastAPI

Airpy – Air Pollution Data Cleaning Module

A Python package for cleaning and processing CPCB (Central Pollution Control Board) air quality data for official government and research use.

Python PyPI Data Cleaning CPCB Air Quality

Key Highlights:

  • Flexible input — process single files or entire directories in CSV and Excel (XLSX/XLS) formats
  • Auto-detects filename format, extracts metadata, removes outliers and consecutive repeats
  • Unit standardization — converts all nitrogen compounds to µg/m³ with debug-friendly verbose mode

Student Performance Prediction

Machine learning system to classify student performance with end-to-end pipeline and deployment.

Scikit-learn Flask AWS EC2 MongoDB

Key Highlights:

  • Developed ML model following industrial approach with pipelining for efficient data ingestion and prediction workflows
  • Built Flask webapp and deployed on AWS EC2 monolithic server
  • Integrated user authentication and secure data storage using MongoDB

Education

B.Tech in Computer Science and Technology

University of Engineering and Management, Kolkata
2020 - 2024 CGPA: 8.8/10

Publications

Assessing Performance of The Jal Jeevan Mission Using a Geospatial Decision Support System

Conference: Center for Public Policy, IIM Bangalore

View Publication

Central Bank Digital Currency (CBDC) and Application Development

Authors: Dr. Prasenjit Das, Magapu Shanmukh, Chandan Kumar, Shubham Kulkarni

DOI: 10.22214/ijraset.2023.57703

View Publication

Get in Touch

I'm always interested in hearing about new opportunities, collaborations, or just connecting with fellow AI/ML enthusiasts. Feel free to reach out!