Skip to content
View RedHeartSecretMan's full-sized avatar
💭
I may be slow to respond.
💭
I may be slow to respond.
13 followers · 33 following
  • 中国
  • 11:51 (UTC -12:00)

Achievements

Achievement: Pull SharkAchievement: Pair Extraordinaire

Achievements

Achievement: Pull SharkAchievement: Pair Extraordinaire

Block or report RedHeartSecretMan

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
RedHeartSecretMan/README.md

Wang Hao

What · Why · How

把复杂的 AI 系统做得可靠、可解释、可复现。
Building trustworthy AI systems, from inference runtimes to evidence-first agent workflows.

Personal Site · 3D Résumé


Rust Python CUDA vLLM Agent Workflows

Focus

What: LLM serving, inference engines, and agent workflows.

Why: Complex AI systems should be trustworthy, inspectable, and reproducible.

How: Rust, Python, CUDA, pinned dependencies, explicit contracts, and evidence-backed verification.

Projects

vllm-dsf-rearguard

Serves DeepSeek-V4.1-Flash on 8×A100 with a pinned vLLM image and an adaptation layer, plus tools for weight verification, image builds, guarded launch, and functional and performance testing.

Python vLLM A100 Model Serving

vllm-oxide

Rust LLM inference engine built on Candle.
Uses custom CUDA kernels and continuous batching.
Adds prefix caching and a paged KV cache.

Rust CUDA Candle

dag-skill

Coordinates approved Ticket dependency graphs.
Runs one execution agent per Ticket in Git.
Reviews before integration.

Agent Skill Multi-Agent Git

a-share-research-skill

Routes A-share research questions to a deterministic Python CLI whose results record the security, as-of date, sources, and calculation lineage.

Agent Skill Python CLI A-share

agent-recipes

Personal agent rules, Skill sources, and MCP settings kept in one repository and shared across Codex, Claude, DSH, and Cursor.

Agent Rules Skills MCP

Stats

Wang Hao's GitHub statistics Most used languages



3D contribution graph



Contribution snake

What matters. Why it matters. How to solve it reliably.

Pinned Loading

  1. vllm-dsf-rearguard vllm-dsf-rearguard Public

    Pinned DeepSeek-V4.1-Flash-1017 adaptation and evidence-based performance experiments for 8x A100 SXM4 80GB

    Python 3

  2. vllm-oxide vllm-oxide Public

    A Rust inference engine for large language models, built on Candle and custom CUDA kernels.

    Rust 2

  3. a-share-research-skill a-share-research-skill Public

    Evidence-first A-share research Skill with a deterministic Python CLI, explicit provenance, fail-closed identity, and reproducible valuation lineage.

    Python 2

  4. dag-skill dag-skill Public

    Coordinate approved Spec-and-Ticket DAGs with evidence-aware review and bounded rework.

    Python 8 2

  5. agent-recipes agent-recipes Public

    Personal agent rules, Skills and MCP recipes

    Python 1

  6. dsh-plugins dsh-plugins Public

    DeepSeek Harness plugins: a catalog of installable bundles for session titles, and whatever comes next.

    JavaScript 1