My name is Maxim Podgore

I'm a software developer and researcher focused on Artificial Intelligence, with current research in Black-Box Optimization and static analysis tools. I have also worked on Mobile Development, UI/UX, Front-End, Database, and Bioinformatics teams.

Resume available here · GitHub

Work Experience

Team working at desks with laptops — UCSD Information Technology Services
UCSD Information Technology Services team

Machine Learning Engineer / TL January 2024 - Present

UCSD Information Technology Services

  • Developed an AI contract review agent leveraging LiteLLM, LLaMA, and Qwen-Embedding for prompt tuning via user feedback. The agent uses a novel Inverse Reinforcement Learning (IRL) pipeline to suggest edits to legal documents and learn from user feedback, driving a 10% increase in contract settlements per month.
  • Built an AI-powered OCR pipeline to assist with transcript validation via NanoNets, NVLM-D, and Docker for layout processing, OCR, and business logic, automating the processing of 30,000 UCSD student transcripts.
  • Led a team of 4 student engineers to design and deploy ML solutions, mentoring and assisting with technical problems throughout the process.
SingularityNET logo
SingularityNET internship work

Machine Learning Engineer Intern July 2025 - September 2025

SingularityNET

  • Resolved knowledge conflicts in a Graph RAG system by integrating timestamps and confidence-based sampling into Neo4j-based graph construction, improving consistency and reliability of retrieved information by 30%.
  • Refactored a cutting-edge benchmark to extend to graph-RAG systems using qwikidata and pandas, testing knowledge recall and ripple effects caused by introducing counterfactual knowledge triplets to the system.
Researcher viewing static analysis visualizations — Stanford Novel Computing Lab
Stanford Novel Computing Systems Lab research

Artificial Intelligence Researcher May 2024 - Present

Stanford Novel Computing System’s Lab

  • Built a C++ static analysis tool for the dReal SMT solver to address floating-point soundness issues, implementing a graph-based LLVM AST solver to validate rounding mode constraints, which detected 200+ problematic lines.