Blog & Notes

Occasional notes on LLMs, autonomous software engineering, research engineering, and things I'm learning.

• A-ProS: Multi-Model LLM Feedback Loops for Reliable Autonomous ProgrammingMay 20, 2026

  A-ProS (TOSEM 2026) is an agentic framework that separates solution generation (GPT-4, GPT-5) from specialized debugging feedback (DeepSeek-R1, Llama-3.3, Codestral). Stateful multi-model feedback loops deliver 2.2-2.3× greater gains than stateless baselines and reduce error repetition by 2.9-3.5× on 367 ICPC and Codeforces problems.

• Hello, world - and what I plan to write hereMay 18, 2026

  A short introduction to this site and the kinds of notes I want to write - mostly on LLMs for code, agentic systems, and research engineering.

• LLM-ProS: Benchmarking GPT-4o, Mistral, Llama, and OpenAI o1 on ICPC World Finals ProblemsMar 10, 2025

  A rigorous evaluation of five large language models — GPT-4o, Mistral Large, Llama-3.1-405B, o1-mini, and o1-preview — on 166 ICPC World Finals problems (2011-2024). Chain-of-thought reasoning models substantially outperform general-purpose LLMs on hard competitive programming tasks.