Tafazzul Nadeem
- Generative AI Researcher at IIT Kanpur
- MS in Computer Science & Engineering
- Focus on Agentic AI
| GitHub |
I am an MS-Research student in Computer Science & Engineering at the Indian Institute of Technology, Kanpur, with a focus on deep learning and applied AI. My current research involves exploring technical image understanding via VLMs. I have previously worked on Emotion Detection using transformer based models. I have built intelligent AI systems combining research and practical deployment. Key projects include Agentic RAG chatbot, Agentic tool for automating calendar events from emails. I am particularly interested in building Agentic AI Deep Research systems employing Context Engineering, RAG and tool calling.
Core Technical Skills
| Category | Skills |
|---|---|
| Programming Languages | Python, C |
| Frameworks and Libraries | Huggingface, Langchain, Langgraph, MCP, pytorch |
| Relevant Coursework | Introduction to Machine Learning, Statistical NLP, Deep Reinforcement Learning, Parallel Computing |
Research Experience & Publications
1. Investigations into Technical Image Understanding by VLM
Research Project | Guide: Dr. Ashutosh Modi | Apr-Aug’25
- Objective: Assessing and benchmarking the capabilities of Vision-Language Models (VLMs) in the domain of technical image understanding.
- Approach: Generated synthetic image–code–summary datasets across 8 diagram types using Mermaid code. Evaluated diverse VLMs on image-to-code, summary-to-code, enhancement, and hand-drawn diagrams.
- Results: Finetuning improved the average ROUGE-L performance of Llama3.2-11B-instruct by 2.13x on the Image-2-code task.
- Publication Status: Research paper under review (as of Sep-25) in AAAI-2026 (A* conference). Next step involves Agentic modeling.
2. Text Based Emotion Detection: SemEval 2025
Course Project CS779-NLP | Guide: Dr. Ashutosh Modi | Sep’24-Feb’25
- Objective: Performed multi-label classification of perceived emotion (Anger, Fear, Joy, Sadness, Surprise) of a speaker’s text.
- Approach: Created an ensemble of pre-trained models (RoBERTa and DeBERTa based) trained on diverse datasets and fully fine-tuned it with added classifier layers.
- Results: Achieved an F1 score of 0.734 with only a 0.7B parameter model, notably outperforming the leader’s model which utilized 665B+gpt4o.
- Publication Status: System paper published in Exploration Lab IITK at SemEval-2025 Task 11: Bridging the Gap in Text-Based Emotion Detection
Key Projects
1. medScan-BotAgent - Agentic AI chatbot with RAG
Link: nadeemtafazzul/medScan-BotAgent
- Developed an Agentic AI chatbot giving information about rates and offers using RAG (Retrieval-Augmented Generation), powered by OpenAI’s GPT-4o-mini, for a medical diagnostics lab website.
- The project was prototyped in Gradio and deployed on Huggingface.
- Performance: Obtained 98% score on manually curated 10 diverse questions on information retrieval from the knowledge base (2% hallucination).
2. whiner - Agentic AI system for creating Google Calendar events from emails
Link: Tafazzul-Nadeem/whiner
- Designed an Agentic AI system using LangGraph powered by Gemma that reads email inboxes and automatically creates calendar events with deadlines and details.
- Ensured privacy by using codeword substitution of private data before processing.
- Performance: Achieved 96.4% score on private info cleaning and 70% on correct event creation in human evaluation of 10 consecutive e-mails.
3. Deep Research with LangGraph (Online Course: Langchain Academy)
Link: Tafazzul-Nadeem/deepResearch-LangGraph
- Completed LangChain Academy’s Deep Research course on building agentic AI systems for deep research applications.
- Techniques used included context engineering, tool calling, and MCP servers.
- Achievement: The system was ranked 6th on the deepResearch bench on Huggingface.
4. DIAYN Paper Implementation (Deep Reinforcement Learning)
Course Project CS780-Deep Reinforcement Learning | Link: Tafazzul-Nadeem/DIAYN-deepRL
- Implemented the DIAYN (Diversity is All You Need, Eysenbach et al.) framework across multiple Gymnasium environments.
- Trained agents to learn skills without a reward function. Trained for 50 skills on 6 different environments.
Scholastic Achievements
- Secured All India Rank (AIR) of 200 in GATE-CSE 2023.
- Dual Rank Holder in GATE-CSE (AIR-51) in 2017 and GATE-EE (AIR-612) in 2016.
- Awarded the ”Build India Scholarship” from L&T Construction (40 selected all India).
Work Experience (Prior to Research)
National Aluminium Company Ltd (Central Govt. PSU): Officer Grade
Apr’17 to Dec’23
- Joined as a Graduate Engineer Trainee (GET) through GATE-2016.
- Worked in the Electrical Maintenance Dept of Nalco Smelter Plant. Responsibilities included managing day-to-day maintenance of power and automation equipment, as well as executing upgradation projects.
- Resigned as Deputy Manager (Electrical).
- Recognition: Received 10 ‘Sarjana Suggestion Awards’ and the Best Energy Saving Project of 2022.
Certification
- Lean Six Sigma Green Belt Certification (Nov’22) from Q3Edge Consulting Pvt Ltd, obtained during an internal training programme in Nalco.
Extra Curriculars
Outside of academics and research, I’ve always enjoyed sports — especially roller hockey and table tennis. I played roller hockey at the national and zonal levels during my school years, which sparked my love for team sports and competition. Later, I actively participated in table tennis, playing in district and college tournaments, and it remains one of my favorite ways to unwind.