Skip to content
View RenataLi's full-sized avatar
💭
👩‍💻👾
💭
👩‍💻👾
1 follower · 0 following

Achievements

Achievement: YOLOAchievement: Pull Shark

Achievements

Achievement: YOLOAchievement: Pull Shark

Block or report RenataLi

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.

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
RenataLi/README.md

ML/LLM engineering, NLP, and reproducible AI evaluation

I am an ML/LLM engineer and data scientist building reliable language-model systems—from post-training and cross-lingual representation analysis to RAG, information retrieval, and tool-calling agents.

My work emphasizes controlled experiments, exact grading, statistical uncertainty, failure analysis, and artifacts that others can rerun.

Selected work

  • Agent Reliability Lab — a deterministic, fault-injected benchmark for tool-calling agents with typed tools, policy-gated writes, replayable scenarios, and exact outcome grading.
  • LatentSuff — a preregistered cross-lingual study of evidence-sufficiency signals in open-weight LLM representations across six languages and 29,206 paired QA constructions.
  • LIMIT+ follow-up — an independent information-retrieval study separating BM25 candidate coverage from neural and symbolic constraint ranking.
  • Reasoning Post-Training Lab — audit-first QLoRA SFT and DPO experiments with matched controls, locked selection, and hash-bound provenance.

Focus: ML/LLM systems · NLP and cross-lingual analysis · RAG, information retrieval and reranking · agent reliability · post-training

Tools: Python · PyTorch · Transformers · vLLM · CUDA · pytest · Docker

Pinned Loading

  1. open-weight-agent-reliability-lab open-weight-agent-reliability-lab Public

    Deterministic fault-injection benchmark for open-weight tool-calling agents with typed tools, policy-gated writes, exact outcome grading, and reproducible local-GPU studies.

    Python

  2. latent-sufficiency-lab latent-sufficiency-lab Public

    Preregistered cross-lingual evidence-sufficiency probing in Qwen3 — 6 languages, 29,206 paired QA constructions, frozen English probes, and a sealed one-shot evaluation.

    Python

  3. limit-plus-independent-followup limit-plus-independent-followup Public

    Independent reproducible LIMIT+ study separating candidate coverage from constraint ranking with BM25, Qwen3-Reranker-4B, and a symbolic upper bound.

    Python

  4. open-weight-reasoning-lab open-weight-reasoning-lab Public

    Reproducible QLoRA SFT and DPO experiments for open-weight LLM reasoning, with locked selection, matched controls, and audit-grade provenance.

    Python