Skip to content
View Shumatsurontek's full-sized avatar
💭
Spelunking
💭
Spelunking

Block or report Shumatsurontek

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. 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
Shumatsurontek/README.md

Arthur Edmond — AI / LLM Engineer

Building production-grade LLM systems for real-world automation MS Electrical Engineering, INSA Lyon · Paris, France

I design, evaluate, and ship LLM-powered systems in production, with a focus on agent architectures, retrieval pipelines, model evaluation, and cost-efficient AI infrastructure.

My work sits at the intersection of applied LLM engineering and research: turning experimental ideas into reliable systems that can handle real user workflows, production constraints, and measurable quality requirements.

What I work on

Agents          Multi-agent workflows, LangGraph orchestration, dynamic model routing, lazy skill loading
Retrieval       Hybrid search, BM25/vector fusion, reranking, confidence scoring, OpenSearch indexing
Evaluation      LLM-as-judge, golden datasets, offline benchmarks, cross-model comparison
Infrastructure  Prompt caching, embedding deduplication, async workers, production observability

Selected research & experiments

Project Focus
neo-deep-agent-lab Deep agent SQL lab with serverless deployment on Modal
neo-neo-sn62 Bittensor SN62 mining and validation for decentralized AI systems
steering-research Research on activation steering in instruction-tuned language models
Gptq-Babai-Quantization Visualization of GPTQ and Babai lattice reduction for model quantization
sparse-moe-torch Sparse Mixture-of-Experts routing and load-balancing experiments

Technical interests

  • LLM systems — agents, tool use, retrieval-augmented generation, model routing
  • Evaluation — LLM-as-judge, benchmark design, regression testing, production quality gates
  • Mechanistic interpretability — activation steering, feature circuits, sparse probing
  • Quantization — GPTQ, lattice reduction, mixed-precision strategies
  • Mixture of Experts — routing policies, expert load balancing, capacity factors
  • Information retrieval — hybrid search, reranking, reciprocal rank fusion, learned sparse retrieval
  • Reinforcement learning — policy gradients, RLHF/DPO, reward modeling

Stack

Python FastAPI LangGraph PostgreSQL pgvector OpenSearch AWS Bedrock Cohere Celery SQLAlchemy


Paris, France

Pinned Loading

  1. obsidian-vault obsidian-vault Public

    MCP server + autonomous agent that links, tags, and curates an Obsidian vault on a schedule

    Python 1

  2. steering-research steering-research Public

    A deep dive into where, how, and why activation steering works (or doesn't) on instruction-tuned LLMs

    Python 2

  3. neo-deep-agent-lab neo-deep-agent-lab Public

    deepagent-sql-lab with modal implementation

    Python 1

  4. local-agent local-agent Public

    Python 1