Ann Arbor, Michigan
I build Python and TypeScript applications for agent workflows, hybrid retrieval, and engineering automation. My work connects model behavior to typed APIs, review controls, and usable interfaces.
Open to AI full-stack, platform, and product engineering roles
Python / TypeScript / Applied AI
Applications
A résumé application and a household pantry assistant, with hosted entry points below.
Evidence-based application workflow
Finds jobs, checks tailored claims against profile facts, and tracks applications in separate user workspaces.
Shared pantry and meal planning
A Telegram bot and Mini App for receipt imports, shared household inventories, expiry reminders, and meal planning.
Selected work9 projects
Applications, developer tools, and enterprise systems. Open-source contributions are identified separately from original projects.
EVIDENCE CHECKS · VISUAL PDF REVIEW
Finds jobs and tailors résumés against profile evidence. Gmail, Outlook, and IMAP connections propose tracking updates for review. External agents prepare applications with revision checks on answers and verified attachments. Typst exports include evidence-backed sections, page-fit checks, and visual layout review.
SHARED PANTRY · TELEGRAM, SLACK & WHATSAPP
A Telegram bot and multilingual Mini App for receipt scanning, shared pantry records, expiry reminders, and meal plans. Purchases can be grouped by store, batch changes need confirmation, and past choices inform food preferences. Self-hosted deployments also connect Slack and WhatsApp with the same household permissions.
SAVED REVIEW PLANS · QUARANTINE
Uses FFmpeg and cached fingerprints to find duplicate videos and shared clips. Review matches in a browser, reopen saved plans, and track scan and removal progress. Decisions about a file carry across clip groups. Approved removals move to quarantine after the tool checks coverage by retained videos.
TASK SCHEDULING · SOURCE PREVIEWS
A customized LangChain console with authentication, usage budgets, and scheduled tasks. Chat resumes from current checkpoints and keeps each approval tied to the interrupt that requested it. File mentions, queued messages, and previews of artifacts, LLM Wiki search results, page metadata, and source images support ongoing work.
REVISION CHECKS · VERIFIED ALM WRITES
Requirement Buddy is an engineering dashboard for source research, requirements drafting, independent review, and verification. Its SDK workflows can resume after interruption and check source revisions before reuse. Separate Jira, Confluence, and Polarion MCP servers handle reviewed publication batches and verify test-execution writes by reading the results back.
SCHEDULED TASKS · ENTERPRISE RETRIEVAL
A LangGraph supervisor runs enterprise tools, hybrid retrieval, and isolated code analysis under shared permissions and budgets. Users can schedule their own tasks or enable a Teams bridge. Reviewed writes recheck their targets; SharePoint downloads and versioned private skills preserve source context.
RETRYABLE EXTRACTION · BUILD TRACEABILITY
Parses release emails into typed records and Jenkins build requests. Failed extractions are kept for retry. The containerized reviewer serves its browser dependencies locally and requires HTTPS for remote Graph token updates by default. Run summaries report failures and distinguish queued requests from completed builds.
BOUNDED CONCURRENCY · RECOVERABLE EXPORTS
Python clients and workflows connect Jira, Polarion, Confluence, Teams, and email through typed interfaces. Bulk retrieval adjusts concurrency within operator limits, keeps input order, and rejects incomplete Jira searches. Polarion exports retry dropped connections and deferred timeouts. SOAP and REST clients share test-result image handling.
INDEPENDENT REVIEW · APPROVAL RECEIPTS
Copilot or Codex workers research sources, draft pages, and independently review the text and locally rendered diagrams. A Python host validates files passed between workers, generates Confluence XML, and checks page versions. Publication requires an approved content digest. Optional human review adds previews and approval receipts.
Engineering experience5 tracks
Agent orchestration, model integration, and the controls around software that acts on a user's behalf.
Agent platform engineering
The same permissions and budgets apply to chat, each user's scheduled tasks, and the optional Teams bridge. Writes to Jira, Confluence, and Polarion require administrator review, a fresh target check, and a persistent record of the claimed change.
Multi-provider LLM interoperability
Pantry Manager supports Anthropic, OpenAI, Gemini, and DeepSeek. Receipt parsing falls back to a vision-capable provider when the selected text model cannot process images.
View related project (opens in a new tab)Trustworthy AI output
Résumé checks cover source citations, skill names, and numeric evidence, including optional sections. Typst checks page fit, while reviewers inspect rendered pages for layout. Each generated artifact records the verified skill procedure used to produce it.
View related project (opens in a new tab)Real-time streaming interfaces
The customized LangGraph console streams tokens and state and resumes from current checkpoints. Approvals stay tied to the interrupts that requested them. Users can manage scheduled tasks, mention files, and preview source images from chat to follow background work and inspect its evidence.
View related project (opens in a new tab)Technical leadership
Personally triaged 267 issues and directed more than 430 across nine ADAS programs, coordinating algorithm, integration, testing, and data-mining teams.
Résumé
Experience, education, and publications.
Aptiv Corporation Troy, Michigan
University of Michigan College of Engineering Ann Arbor, Michigan
Varian Medical Systems, a Siemens Healthineers company Palo Alto, California
Contemporary Intelligent Manufacturing Co. Ltd. Ningde, China
Shanghai MAXIEYE Technology Co. Ltd. Shanghai, China
University of Michigan — Ann Arbor
Automotive Engineering concentration · GPA 3.9
Shanghai Jiao Tong University
GPA 3.6
M. Liu, F. Zhao, J. Wu, W. Fan, J. Zhang, and D. Hung
SAE WCX
J. Wu, J. Wang, Y. Yang, Z. Zeng, D. Lu, and L. Lian
OCEANS
Engineering approach
I design the boundaries around model calls: what data enters, what actions are allowed, and how people review the result.
Typed contracts and fact locks.
Roles, approvals, budgets, and audit trails.
Retrieval, agents, APIs, interface, and delivery.
Toolkit
Contact
Open to AI full-stack, platform, and product engineering roles
Contact me about AI application engineering, agent platforms, or full-stack product development.