Skip to content
View estelledc's full-sized avatar

Highlights

  • Pro

Block or report estelledc

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
estelledc/README.md

Jason Xun

AI Application Engineer · Reliable agents, RAG/evals · Web/iOS delivery

我做可运行、可评测、可恢复的 AI 应用。不只把模型接进产品,也把工具边界、失败语义、证据回放和人工责任写进系统:它做了什么、为什么失败、怎样重放、何时交还给人,都应该说得清。

English summary: I build reliable AI applications with bounded tools, replayable evaluation, explicit failure semantics, and human ownership across Web and iOS delivery.

Work · Résumé · About

Selected evidence

  • TraceFetchv1.0 · Public source 把搜索候选与可信证据分开,产出带原文、锚点、收据和 SHA-256 的可验证 evidence bundle;不判断来源真假,也不绕过访问控制。
  • BJ-Palv6.29 · Public source 用 durable jobs、有界模型执行和可复算评测承载短途规划;公开证据不外推生产容量或用户效果。
  • Tencent/WeKnora PR #1785Merged OSS contribution 过滤 Excel 中的 DISPIMG / IMAGE 图像函数串,并补齐 4 条回归用例。
  • web-plan-execute0.9.0-rc.1 · Public RC 把规划、独立复审、执行、交接和验证收敛为一份可恢复的 living ExecPlan;当前仍是 RC。

产品案例、私有项目与 iOS 经历统一由 Work 页面承载。

How I work

  • 失败关闭 — 权限、证据或边界不满足就停止并暴露原因,不静默降级成“成功”。
  • 可重放评测 — 保存输入、来源、版本、决策和收据,让结果能复算、对比和回归。
  • 诚实边界 — Release、RC 与测试收据只证明对应版本和范围,不等于生产 SLA、规模化运行或真实用户效果;AI 的越界点和人的最终责任必须明确。

Pinned Loading

  1. tracefetch tracefetch Public

    Safe search routing and provenance-bearing evidence bundles for AI research workflows.

    Python

  2. bj-pal bj-pal Public

    Evidence-driven Beijing short-trip planning agent with durable jobs, bounded LLM execution, and reproducible evaluation

    Python

  3. web-plan-execute web-plan-execute Public

    Experimental lab for reliable cross-surface planning, execution, handoff, and evaluation.

    Python

  4. Tencent/WeKnora Tencent/WeKnora Public

    Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.

    Go 19k 2.7k

  5. system-design-workbench system-design-workbench Public

    Offline-first system design practice, deterministic analysis, scoring, and evidence export

    JavaScript

  6. langchain-langgraph-langsmith-tutorial langchain-langgraph-langsmith-tutorial Public

    LangChain + LangGraph + LangSmith 入门教程实践

    Python 18