Xinwei Xiong

Agent Systems Engineer

Recovery / Tool Policy / Evidence and Evaluation

I design bounded, recoverable agent execution with tool policy and source evidence, using user scenarios to define acceptance and regression cases.

Engineering Signals

Agentry 191 tests / race detector / 63.2% coverage / stdio MCP
StripSearch SQLite recovery / Source withdrawal / Four export formats
OpenIM #1 contributor / 358 contributions / 11-service Helm / API + E2E CI

Experience

Telepace AI

Founder / Product Engineer

Owned backend, async pipelines and Kubernetes delivery; stopped after 0 paying users.

Remote

OpenIM

Open-source Intern, then Community Contributor

Ranked #1 with 358 contributions; delivered 11-service Helm, API/E2E CI and contributor governance.

Remote

Selected Systems

Agentry

Runtime Design and Implementation

Jun 2026 · embedded prototype Working prototype

A resumable, auditable agent runtime that unifies loops, tools, sessions, policy and multiple transports.

  • Bounded loop; replayable JSONL sessions; allow/deny/ask; HTTP-tool rechecks for DNS, redirects and private ranges.

StripSearch

Product design and AI-assisted engineering

Sep 2026 - present Workspace live · Research unavailable

Person research with traceable evidence, persistent progress and JSON / Markdown / HTML / PDF exports.

  • SQLite checkpoints and action receipts; source withdrawal invalidates dependent conclusions. Live research is unconfigured.

OpenIM

Open-source Intern / Community Contributor

Public contribution window: Apr 2023 - Apr 2024 Past contribution

Owned deployment, CI and contributor governance in a large Go IM project; reached #1 on GitHub's contributor graph.

  • Delivered 11-service Helm/Helmfile, API + E2E CI, and RFC, DCO, PR-review and release governance.