My Experience

A journey through my academic and professional achievements

Publications

The Last Mile: A Novel, HotSpot Based Distributed Path-Sharing Network for Food Deliveries
IEEE Transactions on Intelligent Transport Systems

DOI: 10.1109/TITS.2024.3465217

Authors: Ashman Mehra, Divyanshu Singh, Vaskar Raychoudhury, Archana Mathur, Snehanshu Saha

IEEE Xplore Google Scholar

Work Experience

Research Intern

CLAN Labs, Purdue University, West Lafayette, IN

May 2025 - Present

  • Working on the development of dynamic obstacle avoidance techniques using uncertainty-based adaptive planning with diffusion models.
  • Exploring multi-modal applications of diffusion models in medical domains.
  • Working under the supervision of Dr. Vaneet Aggarwal.

Undergraduate Researcher

APP Center for AI Research (APPCAIR), Goa, India

December 2023 - Present

  • Developed DeliverAI, a reinforcement learning-based model optimizing food delivery routes.
  • Investigating a novel dynamic ride-sharing system using a multi-agent actor-critic approach with enhancements to minimize detours and optimize rider pickups.
  • Formulating a driver behavior modeling framework to quantify unruliness in traffic scenarios, with plans to adapt it for complex traffic conditions in India.
  • Working under the supervision of Dr. Snehanshu Saha and Dr. Santonu Sarkar.

Research Intern

Digital India Bhashini Division, India

May 2024 - July 2024

  • Contributed to the National Language Mission to advance language technologies for Indian languages.
  • Collaborated with the post-processing team, focusing on Inverse Text Normalization (ITN).
  • Designed a WFST-based model for ITN across diverse Indic languages.
  • Implemented a T5-based model to enhance ITN by introducing punctuation handling.

Teaching Assistant

CS-F425 Deep Learning, BITS Pilani, Goa, India

January 2025 - May 2025

  • Designed and conducted tutorials on deep learning concepts for undergraduate students.

Undergraduate Volunteer

IndoML 2024, Goa, India

December 2024

  • Represented APPCAIR as one of the few undergraduate volunteers.
  • Showcased research at the sponsor stall and facilitated Q&A sessions during the conference.

Research Projects

Graph Diffusion for RNA 3D Folding
BITS Pilani, Goa (March 2025 - Present)

Technologies: Graph Neural Networks, Diffusion Models, Computational Biology

  • Developing a graph diffusion model to predict 3D structures of RNA molecules by leveraging graph neural networks and probabilistic sampling techniques.
  • Exploring the integration of structural constraints and energy-based priors to enhance the accuracy of RNA folding predictions.
  • Evaluating the model's performance on benchmark datasets, focusing on metrics like RMSD and structural fidelity to native conformations.
Driver Behaviour Modeling
BITS Pilani, Goa (December 2024 - Present)

Technologies: Generative AI, Computer Vision

  • Developing a transformer-based encoder-decoder model integrating kinematic data, vectorized maps, and vision-language outputs for real-time anomaly detection in traffic behavior.
  • Designing a driver unruliness quantification framework, capturing risky maneuvers, violations, and aggressive driving patterns.
Altruistic Ride Sharing
BITS Pilani, Goa (August 2024 - Present)

Technologies: Deep Reinforcement Learning, Optimization

  • Developed an altruistic ride-sharing framework where drivers can pick up and drop off riders along their route while minimizing detours.
  • Utilizing an Attention-based Multi-Agent Actor-Critic model with novel policy modifications to enhance cooperative decision-making, optimizing ride allocation while balancing altruism and efficiency in ride-sharing.
LLMExam: AI-Driven Question Paper Generation
BITS Pilani, Goa (April 2025)

Technologies: NLP, Retrieval-Augmented Generation, Generative AI, Multimodal AI, Reinforcement Learning

  • Built a multimodal pipeline using a local LLM (Qwen 2.5) for automated retrieval of textual questions, diagrams, and graphs from PDFs, leveraging vector search for topic-based extraction.
  • Implemented LLM-driven topic tagging with dynamic matching against a vector database for accurate question categorization.
  • Generated curated question papers in Markdown format based on user-defined topics and constraints, preserving textual and visual content.
  • Developing a Reinforcement Learning with Human Feedback (RLHF) framework to evaluate and improve topic tagging accuracy, difficulty assignment, and overall pipeline stability.
DeliverAI
BITS Pilani, Goa (December 2023 - June 2024)

Technologies: Reinforcement Learning, Optimization

  • Proposed DeliverAI, a reinforcement learning-based path-sharing algorithm to optimize food delivery routes by reducing costs and improving efficiency.
  • Modeled the problem as a multi-objective optimization balancing consumer satisfaction and delivery costs.
  • Simulated DeliverAI on real-world Chicago delivery data, demonstrating a 15% reduction in fleet size, 16% reduction in distance traveled, and 50% increase in fleet utilization compared to traditional point-to-point delivery methods.
  • Published in IEEE Transactions on Intelligent Transportation Systems.
IEEE Xplore
Zero-Shot Classification with RoBERTa
BITS Pilani, Goa (November 2024)

Technologies: NLP, Deep Learning, Generative AI

  • Utilized a pre-trained RoBERTa model for zero-shot classification on the AG News dataset using Hugging Face transformers.
  • Optimized label prompts iteratively with generative LLMs (Gemma2-9B, Qwen2.5-32B, Nemotron-70B), boosting accuracy from 48.5% to 82.13% (achieved by Nemotron-70B).
  • Assessed performance using precision, recall, F1-score, and confusion matrices, analyzing common error patterns and label effectiveness.
GitHub

Technical Proficiency

Programming Languages

Python, C++, C, Java, SQL, LaTeX

Software and Tools

PyTorch, TensorFlow, JAX, Hugging Face, Gymnasium, PettingZoo, Anaconda, GitHub, Docker

Interests

Reinforcement Learning, Generative AI, Deep Learning, Natural Language Processing, Optimization

Education

Bachelor of Engineering in Computer Science
Birla Institute of Technology and Science, Pilani, Goa, India

2022 - 2026

Relevant Coursework: Reinforcement Learning, Generative AI, Natural Language Processing, Foundations of Data Science, Object-Oriented Programming, Data Structures and Algorithms, Database Systems, Design and Analysis of Algorithms, Discrete Structures in Computer Science, Logic in Computer Science, Theory of Computation

Online Courses
  • DeepMind x UCL: Deep Learning Lectures
  • DeepMind x UCL: Reinforcement Learning Lectures
  • CS229 Stanford ML
  • CS50's Introduction to AI with Python