Research¶
Interactive Agentic Code Generation for Fine-Tuning LLMs¶
Aug 2025 – Dec 2025
- Built an interactive agentic coder with LangGraph, FastAPI, and RAG to generate and debug code for fine-tuning collaborative LLMs.
- Improved coding speed and accuracy for non-expert users; submitted to NeSy 2026.
Human–AI Trust & Relationship Modeling in Conversational Agents¶
Mar 2025 – Oct 2025
- Designed prompts, an A/B-testing framework, and a full-stack agentic chatbot (FastAPI, LangChain) to assess trust.
- Improved user engagement and task completion among 12 pilot users in Phase 1 via an optimized prompting strategy.
In-Context Learning vs Retrieval for Data-Efficient LLM Reasoning¶
Jan 2024 – Dec 2024
- Proposed a framework to identify and manipulate LLMs’ in-context learning (ICL) mechanisms.
- Achieved up to 90% data efficiency; published at NAACL 2025 (Outstanding Paper Award).
Reasoning over Text that includes Uncertainties using Generative LLMs¶
May 2023 – Jul 2025
- Created a Bayesian-inference dataset; prompt-engineered coding methods improved accuracy by 40%, leading to an AAAI 2025 publication.
- Used LLMs to emulate expert probability judgments, improving Bayesian-network accuracy by 7%; submitted to TMLR.
Developing a Neuro-Symbolic Deep Learning Library for Constraint-Based Learning¶
Jan 2021 – Sept 2025
- Researched and unified constraint-utilization methods into a library; EMNLP 2021 (Demo).
- Built the first benchmark for evaluating neuro-symbolic methods and embeddings; AAAI 2023.
- Architected an interactive deep-learning coding pipeline (GPT-4) that boosted developer speed by 500%; NeSy 2024.
