Skip to content
View Aawegg's full-sized avatar
0 followers · 1 following

Block or report Aawegg

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Aawegg/README.md

I build RL environments, agent benchmarks and the verifiers that grade them, plus the agents and RAG systems themselves. The part of AI I care about most is whether a number is actually true: deterministic checks, decoy paths that must fail, and systems that are hard to fake.

  • 400+ agent benchmark tasks in Harbor format: Terminal-Bench, multi-agent (SwarmBench) and enterprise-tool tasks over MCP
  • Pod lead, LLM evaluation at Turing (Terminal Bench, OS Bench, SWE-Bench)
  • Kaggle: provisional silver (~top 1% of 10,246 teams) in an RL simulation competition, with a C++ simulator and a PPO policy
  • Merged open source: Rhesis #2173, OGX / Llama Stack #6341
class Aaweg:
    role       = "Applied AI Engineer"
    focus      = ["RL environments", "agent benchmarks", "LLM evaluation", "agents"]
    philosophy = "is it actually good, or just demo-good?"
    currently  = "building Harbor tasks with deterministic verifiers"
    stack      = ["Python", "Docker", "Harbor", "pytest", "LangGraph", "C++", "GCP"]

Featured work

Project What it is Try it
nibblecore 4-bit (Q4_0) quantization kernels in C++ / ARM NEON with a GGUF loader · 3.4x faster dequantization than llama.cpp scalar, 4.8x faster SDOT dot product, bit-identical Repo
priorityjudge Multi-pass LLM-as-judge for plans/PRDs with a deterministic citation verifier · 9887/10000, perfect rank correlation with a human grader on a 6-plan calibration set Quickstart
mcp-arena Reliability scorecard for MCP servers (conformance, tool reliability, hygiene) · WIP Repo
swarmbench-codespaces-harness Run Harbor multi-agent benchmark tasks (oracle / single / multi) on free GitHub Codespaces Repo
mosaicmind-rag Multimodal RAG (PDF/image/audio/video) with LangGraph + LlamaIndex + Chroma · FastAPI · MLflow · Docker Live demo

Stack I reach for

Python C++ Docker PyTorch FastAPI LangChain Pydantic GCP Hugging Face GitHub Actions

also: Harbor · Terminal-Bench · SWE-Bench · MCP · LangGraph · LlamaIndex · ChromaDB · MLflow · Airflow · pytest · ARM NEON · PPO

Activity

contribution streak

Reach me

Website LinkedIn Email

Pinned Loading

  1. codeforces-notebook-generator codeforces-notebook-generator Public archive

    Python

  2. Melody-Generation-using-LSTM Melody-Generation-using-LSTM Public

    Implemented Long Short-Term Memory (LSTM) networks for melody generation

    Python

  3. nibblecore nibblecore Public

    4-bit (Q4_0) quantization kernels in C++/ARM NEON with a GGUF loader. 3.4x faster dequant than llama.cpp scalar, 4.8x faster SDOT dot product, bit-identical.

    C++

  4. priorityjudge priorityjudge Public

    Multi-pass agent that grades plans/PRDs on priority-definition quality. Line-cited evidence, deterministic citation verifier, 9887/10000 on a hand-rated calibration set.

    Python

  5. Review-Portal Review-Portal Public archive

    Python

  6. Story-Character-Extractor Story-Character-Extractor Public

    RAG pipeline that extracts structured character profiles, relationships, and roles from short stories using embeddings, a vector DB, and an LLM.

    Python