Project summary

1. Risk-Aware Adaptive Evaluation: Finding High-Impact Failures Under Limited Budgets - This project formulates AI agent evaluation as a sequential allocation problem and aims at identifying a budget friendly method for evaluating while derisking AI agent deployments. The camera ready version is available at: https://openreview.net/pdf?id=F6gtKHvzFo
   Status: Selected at NeurIPS 2026 Workshop on Evaluation of Interactive Agents (IAEval)
2. When Verification Is Not Agreement: Cross-Agent LLM-to-Dafny Formalization - This project aims to correlate model capability and performance when using a cross-agent formalization protocol in a clinical trial rule code generation scenario. This has an immediate impact on AI Safety by exposing how weaker models can pass formal checks while deviating from structural specifications. The camera ready version is available at: https://openreview.net/pdf?id=LW4iZHM6vl
   Status: Selected at NeurIPS 2026 Workshop on AI for Verifiable Coding (VERICODEGEN)

Both the papers have been accepted for the respective workshops as I am seeking funds to present my work at the NeurIPS 2026 workshops. The work will be presented at the workshop venue at Atlanta, GA, USA between Dec 12 and 13, 2026.

What are this project's goals? How will you achieve them?

The project's goal is to present these accepted papers at NeurIPS 2026 workshop on Evaluation of Interactive Agents (IAEval) and AI for Verifiable Coding (VERICODEGEN). During the presentation I will be seeking for diverse feedback from the attendees and experts on these fields and broaden my research depth. I also plan to attend important lectures and sessions during these workshops to get new ideas to dive deeper and deliver more research work in the fields of optimizing AI Agents in verifiable code generation and evaluation. If time permits, I also plan to explore other workshops at the venue and develop new areas of interest.

How will you know it worked?

I plan to pitch our finding to at least 3 research collaborators from labs specializing in the fields of AI evaluation, neuro symbolic method and formal verification to develop future joint projects. I also plan to generate at least 3 research extensions or new projects based on peer feedback and workshop sessions.

How will this funding be used?

This funding will be primarily used for my travel to Altanta for NeurIPS workshop, registration and expenses during the stay. A cost breakdown is give below:

$400 for airfare for rountrip from Dallas to Atlanta, $100 for local transportation, $600 towards accommodation, $100 for poster printing, $300 for workshop registration

Minimum funding requested: $1,000. Maximum funding goal: $1,500.

Who is on your team? What's your track record on similar projects?

I am the first author for both the projects. My co-author for one of the papers is a fellow Grad Student that I met in my class at Georgia Tech. My research background combines Graph Neural Networks, AI Optimization, Neuro Symbolic and Verifiable Coding, machine learning, and AI governance. I am currently pursuing Masters in Computers Science remotely from Georgia Tech, Atlanta.

What are the most likely causes and outcomes if this project fails?

The research work will not be impacted, however, I may not be able to travel to attend the workshops in person. This will cause us to miss critical expert feedback from senior AI safety reviewers, slowing down our timeline for expanding these workshop papers into full, peer-reviewed journal submissions.

How much money have you raised in the last 12 months, and from where?

I have not raised money. My projects were funded by research cyberinfrastructure resources and services provided by the Partnership for an Advanced Computing Environment (PACE) at the Georgia Institute of Technology, Atlanta, Georgia, USA.