AI Forge Projects

This project develops a retrieval-augmented generation (RAG) system that helps the public understand emerging AI legislation. Students build tools that allow users to query proposed or enacted laws, receive accurate summaries, and explore which groups or sectors may be affected. The team designs automated summarization pipelines, implements RAG enhanced with Direct Preference Optimization (DPO), and evaluates the system through metrics and user studies. The project ultimately aims to create transparent, accessible policy-focused AI systems that support informed public engagement and policymaking.

Qualifications:

• Current BS, BSMS, MS student in Computer Science, Data Science, Artificial Intelligence, or a related field at Purdue University.
• Strong background in machine learning and/or natural language processing (NLP).
• Familiarity with RAG-based architecture and DPO.
• Proficiency in Python and experience with ML/NLP libraries.
• Familiarity with Large Language Models (LLMs) and prompting techniques.
• Experience with real-world data analysis, preprocessing, and evaluation metrics.
• Excellent written and verbal communication skills for technical documentation and presentations.
• Ability to work both independently and as part of a collaborative team.

Responsibilities:

• Design and implement RAG models to retrieve and generate responses from legislative databases.
• Implement RAG with DPO
• Develop automated summarization techniques for long-form policy documents.
• Evaluate the system using quantitative metrics and user studies.
• Commit to working at least 10 hours per week. 
• Present progress and findings at weekly team meetings.
• Collaborate with mentors and peers to refine the system and address challenges.
• Generate a final report at the end of the semester.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026
• Based on the project outcome, we will write a paper and submit it to the NLP/CSS/AI conference for publication. 

Mentors: Dr. Tunazzina Islam, Dr. Daniel S. Schiff, Ryan Rittner (MS student, CS), Saahil Mathur (Undergrad, CS), Vedant Thakur (Undergrad, Political Science)

This project examines how large language models may reflect demographic or persuasive bias when generating targeted messages. Instead of producing persuasive content directly, students audit how models respond to prompts containing demographic descriptors, studying differences in framing related to gender, age, and other characteristics. The team develops a bias-auditing framework, analyzes patterns using quantitative methods, and investigates how differences in agency framing vary across groups. The project aims to advance understanding of fairness and bias in automated communication systems.

Qualifications:

• Current BS, BSMS, MS student in Computer Science, Data Science, or Artificial Intelligence at Purdue University.
• Strong background in machine learning and statistics. 
• Strong background with NLP and transformer-based models (e.g., BERT, LLaMA, or RoBERTa). 
• Familiarity with Large Language Models (LLMs) and prompting techniques. 
• Prior experience with bias-related framework development is preferred.
• Experience in running open-source LLMs (e.g., LLaMA, Mistral) on platforms such as Ollama, as well as making API calls to interact with models. 
• Excellent programming skills (e.g., Python). 
• Experience with reading and writing scientific papers (preferred).
• Experience with data analysis, text preprocessing, and evaluation metrics.
• Excellent written and verbal communication skills for technical documentation and presentations.
• Ability to work both independently and as part of a collaborative team.

Responsibilities:

• Design and implement a bias auditing framework to analyze demographic bias in LLM-generated targeted messaging.
• Evaluate the framework using quantitative metrics and user studies.
• Commit to working at least 10 hours per week. 
• Present progress and findings at weekly team meetings.
• Collaborate with mentors and peers to refine the system and address challenges.
• Generate a final report at the end of the semester.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026
• Based on the project outcome, we will write a paper and submit it to the NLP/CSS/AI conference for publication. 

Mentor: Dr. Tunazzina Islam

This project addresses the challenge of semantic hallucinations in machine-generated formal mathematics. Students build an Automated Theorem Testing Framework (ATTF) in Lean 4 that uses GenAI, property-based testing, and RAG to detect semantic errors before proof search begins. The system automatically generates tests, executes them in a persistent Lean environment, identifies counterexamples, and iteratively prompts the model to refine incorrect theorems. By shifting from “prove-to-verify” to a “test-first” paradigm, the project aims to create a new reliability loop for neural theorem proving and symbolic AI.

Qualifications:

• Current BS, BS/MS, or MS student at Purdue University in Computer Science, Data Science, AI, Mathematics, or a related field.
• Strong programming skills in Python; familiarity with LLMs, APIs, and RAG systems is preferred.
• Interest or background in formal methods, proof assistants, or theorem proving (Lean/Coq/Isabelle).
• Experience with machine learning, formal verification, or natural language processing is a plus.
• Comfortable reading and writing functional code (Lean or Haskell-like syntax).
• Good communication skills for documenting findings and presenting results.
• Ability to work independently and collaboratively within a research team.

Responsibilities:

• Implement GenAI + RAG prompting methods to generate Lean 4 test generators and fix faulty theorems.
• Build and maintain a persistent Lean 4 REPL for dynamic execution and testing.
• Develop lightweight property-based test pipelines based on Plausible for theorem validation.
• Run experiments on standard theorem-proving benchmarks and analyze error patterns.
• Present weekly updates, produce a final report, and create a poster for the Undergraduate Research Expo.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026

Mentors: Dr. Guang Lin, Dr. Yuxiang Peng

This project develops computer vision methods to automatically detect and count grocery items in images or videos. Students explore challenges such as occlusion, cluttered scenes, and variation in packaging, and evaluate model architectures that balance accuracy and efficiency. The project involves building baseline models, testing improved detection and segmentation techniques, and experimenting with vision-language model (VLM)–based approaches. Students contribute to data collection, model implementation, experimentation, and analysis, with the potential to produce publishable research in computer vision.

Qualifications:

• Strong Python programming skills.
• Strong knowledge / common sense  in machine learning, e.g.,  linear algebra and probability.
• Familiarity with deep learning (PyTorch preferred).
• Interest in computer vision + a quicker learner. Prior coursework is helpful but not required. 
• Strong organization and communication skills. We expect students to be able to work effectively in a team.

Responsibilities:

• Assist with data collection/annotation.
• Implement models and run experiments.
• Analyze results and contribute to a short project report.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026

Mentors: Raymond A. Yeh and Ph.D. Student

This project designs a human-in-the-loop (HITL) platform that supports scalable TA assistance for coursework, scheduled for deployment in Spring 2026. The system uses retrieval-augmented generation (RAG) to generate draft answers from course materials, which TAs then verify through an efficient review interface. Students develop both frontend and backend components, build RAG-powered response systems, and design verification workflows that maintain accuracy while reducing TA workload. The project offers hands-on experience in ML-powered educational technology and real-world system deployment.

Qualifications:

• Current BS, BSMS, MS student in Computer Science, Data Science, or Artificial Intelligence at Purdue University.
• Be taking up to 12-credits on other disciplines
• Strong background in machine learning.
• Basic understanding of Large Language Models (LLMs), prompting, and their applications.
• Experience with real-world data analysis, evaluation metrics, and preprocessing pipelines.
• Proficiency in programming (Python required; C++ a plus).
• Excellent written and verbal communication skills for technical documentation and presentations.
• Ability to work both independently and as part of a collaborative team.

Responsibilities:

• Develop and refine a frontend and backend of state-of-the-art RAG models to generate accurate and contextually relevant responses.
• Design and implement mechanisms for efficient TA verification of AI outputs.
• Evaluate system performance using metrics such as accuracy, scalability, and user satisfaction.
• Present progress and findings at weekly team meetings.
• Collaborate with mentors and peers to iterate on the system design.
• Generate a final report at the end of the semester.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026
• Present a poster at the AI Forge + Krenicki Center for Business Analytics and Machine Learning Poster Session March 6th, 2026

Mentors: Bruno Ribeiro, Atharva Thakur

This project builds a high-performance LLM API service that supports text generation and embeddings using frameworks such as vLLM or SGLang. Students design backend infrastructure compatible with OpenAI’s API format, optimize for low latency and high throughput, and integrate monitoring tools for real-time performance analysis. The team implements scalable deployment pipelines, embeds LLMs into the service, and benchmarks efficiency across workloads. The project provides experience in production-grade AI systems and large-scale model deployment.

Qualifications:

• Current enrollment in a BS, MS, or PhD program in Computer Science, Data Science, or a related field.
• Strong programming skills in Python and experience with backend development frameworks (e.g., Flask, FastAPI), and experience with priority queueing and distributed systems.
• Familiarity with machine learning concepts, databases, and LLM inference.
• Excellent problem-solving abilities and attention to detail.
• Ability to work both independently and collaboratively in a team environment.

Desired skills:

• Experience with github, uvicorn, APIs, and linux services
• Knowledge of API design principles and RESTful services.
• Interest in natural language processing (NLP) and large language models.

Responsibilities:

• Design and implement the backend architecture for the LLM API, focusing on scalability and performance.
• Integrate LLMs (e.g., LLaMA, Mistral) into the API using vLLM or SGLang for efficient inference.
• Ensure compatibility with OpenAI’s query format and response structure.
• Optimize the system for low-latency responses and minimal resource consumption.
• Develop embedding methods tailored to educational content (e.g., coursework materials).
• Conduct performance benchmarking and tuning to meet production requirements.
• Collaborate with the AI Forge team to integrate the API into existing systems.
• Present progress and findings at weekly team meetings.
• Generate a final report detailing the architecture, optimizations, and deployment process.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026
• Present a poster at the AI Forge + Krenicki Center for Business Analytics Poster Session, March 6th, 2026.

Mentors: Bruno Ribeiro, Atharva Thakur

This project integrates LLM-powered agents into an open-source logistics simulator to enhance decision-making in complex supply chain environments. Students design agents capable of interpreting natural language instructions, analyzing dynamic conditions, and executing context-appropriate actions within the simulation. The project benchmarks LLM-based agents against rule-based systems and humans, evaluates scalability under increasing complexity, and contributes new capabilities to the simulator. The work bridges AI, logistics, and simulation to support more realistic training environments for decision-making algorithms.

Qualifications:

• Current enrollment in a BS, MS, or PhD program in Computer Science, Data Science, or a related field.
• Strong programming skills in Python and experience with backend development frameworks (e.g., Flask, FastAPI), and experience with priority queueing and distributed systems.
• Familiarity with machine learning concepts, databases, and LLM inference.
• Excellent problem-solving abilities and attention to detail.
• Ability to work both independently and collaboratively in a team environment.

Desired skills:

• Experience with github, uvicorn, APIs, and linux services
• Knowledge of API design principles and RESTful services.
• Interest in natural language processing (NLP) and large language models.

Responsibilities:

• Design and implement the backend architecture for the LLM API, focusing on scalability and performance.
• Integrate LLMs (e.g., LLaMA, Mistral) into the API using vLLM or SGLang for efficient inference.
• Ensure compatibility with OpenAI’s query format and response structure.
• Optimize the system for low-latency responses and minimal resource consumption.
• Develop embedding methods tailored to educational content (e.g., coursework materials).
• Conduct performance benchmarking and tuning to meet production requirements.
• Collaborate with the AI Forge team to integrate the API into existing systems.
• Present progress and findings at weekly team meetings.
• Generate a final report detailing the architecture, optimizations, and deployment process.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026
• Present a poster at the AI Forge + Krenicki Center for Business Analytics Poster Session, March 6th, 2026.

Mentors: Bruno Ribeiro, Gabriel Buginga, Mohit Tawarmalani (Business)
AI Forge & Krenicki Center Project

This project develops a new open-source logistics simulator and an accompanying AI framework for neural algorithmic reasoning (NAR). Students design core simulator architecture, implement discrete-event logistics features, and integrate an LLM assistant that enables natural-language scenario creation. The simulator serves both as a realistic environment for training AI models and as an exploratory tool for analyzing complex supply chain behavior. The project combines software engineering, optimization, and machine learning to build an extensible platform for advanced logistics research.

Qualifications:

• Current BS, BSMS, or MS student in Computer Science, Data Science, Artificial Intelligence, or a related field at Purdue University.
• Be taking up to 12-credits on other disciplines
• Strong background in machine learning, statistics, and deep learning, with hands-on experience in PyTorch.
• Proficiency in Python programming and experience with object-oriented design principles.
• Solid foundation in probability and optimization algorithms.
• Experience with discrete-event simulation (e.g. SimPy) is highly desirable. Familiarity with Graph Neural Networks (GNNs) or algorithmic reasoning is a plus. 
• Basic understanding of Large Language Models (LLMs), API use, and prompting (desirable).
• Excellent written and verbal communication skills for technical documentation and presentations.
• Ability to work both independently and as part of a collaborative team.

Responsibilities:

• Modify, experiment, and improve with the SimPy-based discrete-event logistics simulator.
• Develop and implement Neural Algorithmic Reasoning and Graph Neural Network models using PyTorch.
• Design and run simulation experiments to evaluate the effectiveness and robustness of the AI framework.
• Present progress and findings at weekly team meetings.
• Collaborate with mentors and peers to refine methodologies and address research challenges.
• Generate a final report at the end of the semester.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026
• Present a poster at the AI Forge + Krenicki Center for Business Analytics and Machine Learning Poster Session March 6th, 2026

Mentors: Bruno Ribeiro, Gabriel Buginga, Mohit Tawarmalani (Business)

This project explores how generative AI—particularly large language models—can enhance causal discovery algorithms by incorporating broad world knowledge. Students investigate how LLMs can supply commonsense or contextual causal cues that are missing from purely data-driven methods, improving accuracy when data is sparse or ambiguous. The team designs hybrid causal discovery pipelines, evaluates model performance on benchmark datasets, and studies how textual knowledge can be integrated with statistical structure-learning techniques. The project aims to advance next-generation causal inference tools that combine symbolic, statistical, and generative intelligence.

Qualifications:

• Current BS, BSMS, or MS student in Computer Science, Data Science, Artificial Intelligence, or a related field at Purdue University.
• Be taking up to 12-credits on other disciplines
• Strong background in machine learning, statistics, and deep learning, with hands-on experience in PyTorch.
• Basic understanding of Large Language Models (LLMs), API use, and prompting (desirable).
• Ideally, some experience with causal inference methods or graph-based models (desirable but not necessary).
• Solid experience with real-world data analysis and preprocessing pipelines.
• Proficiency in programming (Python required).
• Excellent written and verbal communication skills for technical documentation and presentations.
• Ability to work both independently and as part of a collaborative team.

Responsibilities:

• Modify and experiment with state-of-the-art causal discovery algorithms (e.g., NOTEARS, DAGMA, DAGPA) using PyTorch.
• Develop techniques to incorporate LLM-generated knowledge into causal inference pipelines.
• Evaluate the framework’s effectiveness using metrics such as causal graph accuracy and robustness.
• Present progress and findings at weekly team meetings.
• Generate a final report.
• Collaborate with mentors and peers to refine methodologies and address research challenges.
• Generate a final report at the end of the semester.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026
• Present a poster at the AI Forge + Krenicki Center for Business Analytics and Machine Learning Poster Session March 6th, 2026

Mentors: Dr. Bruno Ribeiro, Jincheng Zhou (Ph.D. student, CS), Dr. Murat Kocaoglu (Johns Hopkins University)

This project investigates how graph neural networks and neural algorithmic reasoning can accelerate large-scale constrained optimization. Students develop machine learning models that approximate or guide the solution of Mixed-Integer Linear Programs (MILPs), study how graph structure maps onto optimization constraints, and evaluate architectures across diverse benchmark problems. The work includes model design, experimental benchmarking, and integration with traditional optimization tools. The goal is to create hybrid AI–optimization approaches that improve efficiency and scalability in logistics, chemistry, and engineering applications.

Qualifications:

• Currently pursuing a BS or MS in Computer Science, Mathematics, Engineering, or related field.
• Strong foundation in machine learning, optimization, and linear algebra.
• Proficiency in Python and PyTorch.
• Experience with computational optimization tools (e.g., Gurobi, CPLEX) is a plus.
• Excellent problem-solving skills and ability to work both independently and in teams.
• Strong communication skills for presenting complex ideas clearly.

Responsibilities:

• Design and implement novel graph network architectures for optimization tasks.
• Conduct experiments to validate the performance of proposed methods.
• Collaborate with team members to troubleshoot and refine algorithms.
• Present progress and findings at weekly team meetings.
• Generate a final report.
• Collaborate with mentors and peers to refine methodologies and address research challenges.
• Generate a final report at the end of the semester and prepare a manuscript for academic publication.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026
• Present a poster at the AI Forge + Krenicki Center for Business Analytics and Machine Learning Poster Session March 6th, 2026

Mentors: Dr. Bruno Ribeiro, Gabriel Buginga (Ph.D. student, CS), Dr. Mohit Tawarmalani (Krannert School of Management), Dr. Can Li (Chemistry)

This project builds an LLM-driven agent inside Minecraft that supports K–12 GenAI literacy through interactive, game-based learning. Students design agents that interpret natural-language input, perform in-game actions via tool calling, and provide guided educational support on topics such as creativity, problem solving, and foundational AI concepts. The work includes dialogue design, safety mechanisms, user-testing protocols, and environment integration. The project aims to create an open educational platform that demonstrates how generative AI can facilitate hands-on, playful learning experiences.

Qualifications:

• Current BS, BSMS, or MS student in Computer Science, Data Science, Artificial Intelligence, or a related field at Purdue University.
• Strong foundation in programming.
• Familiarity with Linux environments, API integration, and object-oriented programming in Python.
• Prior experience with Minecraft playing and LLM prompt.
• Ownership of a licensed copy of Minecraft Java Edition 1.21+ with an active account.
• Ability to work independently and collaboratively in a fast-paced research environment.

Responsibilities:

• Design Educational GenAI Experiences:
  • Collaborate with mentors to create engaging, pedagogically sound interactions within Minecraft.
• Implement Agent Logic and Safety Modules:
  • Use LangChain for tool calling with the Minecraft Java API.
  • Develop a RAG (Retrieval-Augmented Generation)-based system for Q&A.
  • Integrate safety modules to ensure age-appropriate, ethical AI behavior.
• Modify and Extend Existing Agents:
  • Adapt an existing agent framework to meet project goals, ensuring scalability and modularity.
• Conduct User Studies and Evaluate Outcomes:
  • Design and execute experiments to measure learning gains and prompt quality.
  • Analyze data and draw actionable conclusions for iterative improvements.
• Communicate Progress and Results:
  • Present findings at weekly team meetings.
  • Present a poster at the AI Forge + Krenicki Center for Business Analytics and Machine Learning Poster Session March 6th, 2026

Mentor: Dr. Bruno Ribeiro

The semiconductor industry is racing to meet the ever‑growing demand for bandwidth‑intensive applications (AI, high‑performance computing, data‑center workloads). High‑Bandwidth Memory (HBM) is a key component, but its performance is limited by the material properties of the underlying semiconductors.

In collaboration with Kepler Computing, we are building a closed‑loop AI platform that accelerates the discovery of novel materials optimized for high performance memory and logic, lower leakage, and improved thermal stability, key enablers for the next generation of memory and logic modules.

The work will be at the intersection of machine learning, graph neural networks, large‑language‑model (LLM) prompting, and materials science, contributing directly to a pipeline that can propose, evaluate, and recommend new material candidates in weeks rather than months.

Qualifications:

• Current BS or BSMS student in CS/DS/AI/ECE/MSE at Purdue.
• Strong background in machine learning and graph learning.
• Basic knowledge in LLM API use.
• Candidates with knowledge in material science preferred.
• Solid experience with real-world data analysis.
• Excellent programming skills (e.g., Python, C++).
• Excellent written and verbal communication skills.
• Ability to work independently and as part of a team.

Responsibilities:

• Develop and implement an AI model for materials discovery based on relational deep learning and an LLM RAG system.

Mentor: Dr. Bruno Ribeiro

This project develops an AI agent capable of autonomously exploring datasets, identifying emerging concepts, and synthesizing insights that augment human analysis. Rather than relying solely on LLMs to process raw data, the system combines statistical exploration with LLM-based reasoning to surface explanations, patterns, and hypotheses. Students design modular reasoning components, integrate them into an agentic workflow, and evaluate how well the agent reduces human cognitive load during exploratory data analysis. The project advances interactive, reasoning-centered AI systems for data-intensive tasks.

Qualifications:

• Current BS or BSMS student in CS/DS/AI at Purdue.
• Strong background in machine learning.
• Solid experience with real-world data analysis.
• Excellent programming skills (e.g., Python, C++).
• Excellent written and verbal communication skills.
• Ability to work independently and as part of a team.

Responsibilities:

• Develop and implement reasoning modules for data exploration.
• Integrate modules into the data exploration agent architecture
• Evaluate the impact of the developed modules by running experiments and participating in user studies.
• Present research results at weekly meetings.
• Collaborate with other researchers.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026

Mentors: Dr. Dan Goldwasser, Eylon Caplan (Ph.D. student, CS)

This project applies cutting-edge vision-language models to ecological monitoring and wildlife conservation. Students analyze video and image data from natural habitats to detect species, classify behaviors, and identify environmental changes using multimodal AI tools. The team benchmarks VLMs on ecological tasks, builds reasoning modules for habitat assessment, and collaborates with conservation researchers to validate results. The project supports real-world environmental science and demonstrates how AI can assist in sustainable ecosystem management.

Qualifications:

• Current BS or BSMS student in CS/DS/AI at Purdue.
• Strong background in machine learning.
• Ideally some experience with vision tasks.
• Solid experience with real-world data analysis.
• Excellent programming skills (e.g., Python, C++).
• Excellent written and verbal communication skills.
• Ability to work independently and as part of a team.

Responsibilities:

• Develop and implement a visual analysis framework for video data. 
• Benchmark existing VLMs, potentially use self-supervision approach to improve models’ analysis.
• Create a reasoning framework, utilizing scientific knowledge to detect inconsistencies in the models’ analysis. 
• Experiment and compare different reasoning approaches.
• Evaluate models’ performance on existing datasets.
• Present research results at weekly meetings.
• Collaborate with other researchers, by sharing results and data.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026

Mentors: Dr. Dan Goldwasser, Dr. Tania Chakraborty, Dr. Bryan Pijanowski (Forestry & Natural Resources)

This project investigates how foundation-model architectures can incorporate graph-structured data to improve multimodal reasoning. Students evaluate methods for encoding graphs, integrate these representations with language models, and benchmark performance on tasks such as misinformation detection, stance analysis, and social network reasoning. The work includes dataset design, experimental evaluation, and model analysis. The project aims to advance foundation models that reason effectively across both textual and structural information sources.

Qualifications:

• Current BS or BSMS student in CS/DS/AI at Purdue.
• Strong background in machine learning.
• Ideally some experience with graph NNs
• Solid experience with real-world data analysis.
• Excellent programming skills (e.g., Python, C++).
• Excellent written and verbal communication skills.
• Ability to work independently and as part of a team.

Responsibilities:

• Develop an experimental framework for testing graph models
• Collect and organize relevant training and testing datasets
• Benchmark existing graph foundation models on the datasets
• Implement and train a graph foundation model
• Present research results at weekly meetings.
• Collaborate with other researchers, by sharing results and data.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026

Mentor: Dr. Dan Goldwasser

This project supports the development of AI SHARE, a global portal for studying public attitudes toward artificial intelligence. Students work on data ingestion pipelines, metadata structuring, interactive dashboards, and web components that make global survey data accessible to researchers and policymakers. The team also designs visualization tools for exploring trends in trust, risk perception, and AI governance preferences. The project bridges AI, public opinion research, and policy communication to support evidence-based decision-making.

Qualifications

• Current BS, BSMS, or MS student in Computer Science, Data Science, Artificial Intelligence, or a related field at Purdue University.
• Strong programming skills in Python, R, or JavaScript, with experience using modern web frameworks
• Experience with database design and comfort working with structured datasets.
• Strong communication skills and ability to work collaboratively in an interdisciplinary research team.

Responsibilities

• Implement backend or frontend components of the web portal (API endpoints, database schema, UX features, search tools).
• Apply text processing methods to categorize survey questions and assist in metadata tagging.
• Create interactive data visualizations.
• Present progress and findings at weekly team meetings.
• Collaborate with mentors and peers to refine the web portal and address challenges.
• Generate a final report at the end of the semester.
• Present a poster at the Spring Undergraduate Research Conference on April 14-17, 2026

Mentors: Dr. Tunazzina Islam, Dr. Daniel S. Schiff (Political Science), Beecher Baker (Ph.D. student, CS), Indira Dhananjaya Patil (Ph.D. student, CS), Yunzhe Liu (Undergrad, CS)

This project develops a generative video pipeline that transforms manikin-based medical simulations into photorealistic patient videos. Students train and evaluate diffusion models capable of replicating realistic injuries, physiological signs, and clinical scenarios while avoiding privacy concerns associated with real patient data. The project includes dataset preparation, temporal modeling, evaluation of visual fidelity, and integration of medical expertise. The long-term goal is to create synthetic datasets that enhance the training of emergency-care AI systems.

Qualifications

• Current BS, BSMS, or MS student in Computer Science, Data Science, Artificial Intelligence, or a related field at Purdue.
• Strong background in machine learning and deep learning, ideally some experience with vision tasks.
• Proficiency in Python and experience with ML libraries. 
• Familiarity with Large Language Models (LLMs), Vision Language Models, Gen-AI concepts, and prompting techniques.
• Excellent written and verbal communication skills for technical documentation and presentations.
• Interest in GenAI, video generation, or medical simulation applications.
• Ability to work independently and collaboratively.  

Responsibilities

• Designing, implementing, and optimizing generative AI models for video generation.
• Conducting quantitative experiments and benchmarking different model/prompting strategies.
• Collaborate with mentors and peers to refine the system and address challenges.
• Commit to working at least 10 hours per week. 
• Present progress and findings at weekly lab meetings.  
• Generate a final report at the end of the semester.
• Publish findings in top-tier CV/AI conferences

Mentors: Dr. Juan Wachs, Dr. Masudur Rahman, Yupeng Zhuo (Ph.D. student, CS)

This project studies how vision-language models adapt procedural reasoning when tools are missing or real-world constraints disrupt standard workflows. Students build evaluation pipelines, analyze model behavior on step-by-step tasks, and design prompting or reasoning strategies that improve improvisational performance. The work includes video-frame interpretation, error categorization, and multimodal reasoning experiments. The project aims to enhance the robustness, safety, and adaptability of VLMs in practical environments.

Qualifications

• Current BS or BSMS student in Computer Science, Data Science, or Artificial Intelligence at Purdue.
• Strong background in machine learning, preferably with experience in computer vision or multimodal systems.
• Proficiency in Python and PyTorch is required.
• Familiarity with large language models, vision language models, or prompting methods at a basic technical level.
• Experience with real world and potentially noisy data, such as videos or unstructured annotations, is preferred.
• Strong analytical and communication skills and the ability to work both independently and collaboratively in a research environment.

Responsibilities

• Build a visual analysis and evaluation pipeline for procedural video tasks, including frame selection, object enumeration, and model output processing.
• Benchmark multiple vision language models across different prompting strategies and constraint settings.
• Develop new prompting templates or structured reasoning methods that encourage grounded and context-dependent procedural adaptation.
• Conduct detailed error analysis to understand where models fail, such as hallucinating tools, ignoring visual evidence, or repeating canonical procedures despite missing resources.
• Contribute to the design or refinement of evaluation metrics and assist with internal validation.
• Present progress at weekly meetings and collaborate with researchers by sharing code, experimental results, and insights.

Mentors: Dr. Masudur Rahman, Dr. Juan Wachs, Maxwell Ryan Kawada (Ph.D. student, CS), Pronoma Banerjee (Ph.D. student, CS)

This project investigates how vision-language models (VLMs) can provide reward signals for reinforcement learning (RL) agents when traditional reward functions are difficult to define. Students design experiments where VLMs score observations or sequences of actions against natural-language goals, generating semantic rewards for RL training. The team evaluates how different RL algorithms, such as REINFORCE, PPO, and DQN, respond to VLM-based rewards, analyzes factors influencing reward reliability, and compares inferred rewards to ground-truth performance. The project aims to advance understanding of reward design, improve the interpretability of RL training with AI-inferred signals, and develop practical insights for applying VLM rewards in real-world environments.

Responsibilities

• Set up RL tasks using standard environments such as CartPole or simple continuous control tasks.
• Implement reward functions that query a vision language model to score observations or short sequences of observations.
• Train RL agents using algorithms such as REINFORCE, PPO, and DQN, and compare their performance under inferred rewards.
• Create simple evaluation plots, tables, and metrics to assess how consistent the VLM reward is with the true goal of the task.
• Analyze which algorithms succeed, which fail, and why.
• Present progress in weekly meetings and share code and results with other team members.

Qualifications

• Current BS or BS/MS student in Computer Science, Data Science, or Artificial Intelligence at Purdue.
• Basic understanding of reinforcement learning concepts such as policies and rewards.
• Experience with Python and basic ML frameworks such as PyTorch.
• Interest in running experiments, visualizing results, and interpreting model behavior.
• Strong communication skills and willingness to collaborate and iterate on ideas.

Mentors: Dr. Masudur Rahman, Dr. Juan Wachs, Maxwell Ryan Kawada (Ph.D. student, CS)