Anurag Upperwal

About

M.Tech CSE student at IIT Roorkee with experience in generative AI research, RAG pipelines, and synthetic data generation. Interested in building scalable NLP and AI systems for real-world reasoning tasks.

Experience

• Generative AI Research Intern
  Soket AI Labs
  Dec 2025 – Feb 2026
• Teaching Assistant
  Indian Institute of Technology (IIT), Roorkee
  Aug 2025 – Nov 2025
• Role
  Soket AI Labs
  Jun 2025 – Aug 2025
• Teaching Assistant
  Indian Institute of Technology (IIT), Gandhinagar
  Jul 2023 – May 2024

Education

• M.Tech CSE (PG)
  Indian Institute of Technology (IIT), Roorkee
  Computer Science and Engineering · 2024 – 2026
• PG Diploma
  Indian Institute of Technology (IIT), Gandhinagar
  2023 – 2024

Skills

• SQL
• CSS
• FastAPI
• Git
• GitHub
• Python
• Machine Learning
• PyTorch
• Docker
• LangChain
• Object-Oriented Programming
• Natural Language Processing
• C++
• Data Structures & Algorithms
• Faiss
• Database Management Systems
• Huggingface
• Networkx

Projects

• Towards Efficient MoE: CKA-Guided Knowledge Distillation
  CKA, Knowledge Distillation, Mixture-of-Experts

  Designing a novel distillation framework for Mixture-of-Experts with CKA-based loss for teacher-student representations, addressing representation collapse for stable training of compact, domain-adaptable AI.

• Mapping the Mind: Graph - RAG
  Embeddings, Graph Traversal, RAG

  Developed a knowledge graph-driven RAG pipeline with token-aware summarization to improve retrieval and reasoning, achieving 82% faithfulness and 18% hallucination on global reasoning tasks.

• Productivity Agent
  Semantic Retrieval, Memory-Augmented Assistant

  Developed a memory-augmented assistant with semantic retrieval and tool routing for contextual recall of tasks, events, and notes.

• Autograd Enginer
  Python, Dynamic Computation Graphs, Backpropagation, MNIST, MLP

  Built a custom autograd engine with dynamic computation graphs and backpropagation, achieving smooth loss convergence while training a 3-layer MLP on MNIST with about 92% accuracy.

🏆 Achievements & awards

• CodeChef July Long Challenge 2021 Div3 Rank 1026 · 2021

  Secured global rank 1026 out of 22.5k participants.

• International Informatics Olympiad Rank 33 (State) / 667 (Olympiad) · 2016

  Secured state rank 33 and achieved 667 Olympiad rank.

• LeetCode Biweekly Contest 161 Rank 2535

  Secured global rank 2535 out of 31.5k participants.