AI application product / Agent workflow / LLM Eval
ไป็
AI application product, Agent workflow, and LLM Eval with evidence you can inspect
I turn ambiguous AI ideas into workflow, validation, review surfaces, and delivery artifacts. My strongest fit is AI application product, Agent workflow, LLM Eval, AI POC, and developer-tool experience where model behavior has to become usable, observable, and safe to hand off.
Claims become convincing when the evidence remains inspectable.
Observation note
Not a tool list. A recruiter-safe evidence surface for workflow, evaluation, trace, and delivery work that can be inspected without exposing private material.
From ambiguous task to reviewable loop
Multimodal input, visible state, validation checks, and human review
Docked systems
Casefiles arranged as a crewed program rather than a tool dump.
Each mission starts with ambiguity, then gets resolved into an architecture, an operating loop, and a surface that people can keep using.
Calm systems. Real constraints. No theater.
Flight queue
AI Application Product
Shows ability to turn model output into a reviewable operating loop, not just a one-shot generation demo.
Challenge
Meeting material is scattered across audio, video, screenshots, slides, documents, and action items. A useful AI product needs a workflow that keeps source material, processing state, review, and sharing visible.
Architecture
Designed a session-based workflow with capture/upload, source normalization, processing, result review, and redacted sharing. The proof pack groups 42 backend validation checks into acceptance categories that are safe to discuss in interviews.
Ownership
Stack
Outcome
Produced a runnable MVP and public proof case that demonstrate AI workflow product thinking without overstating production maturity.
Define the boundary
Build the loop
Keep it habitable
Public proof
Sanitized case pages that show what each project proves, where the boundaries are, and which AI product roles the evidence supports.
Curvature, rhythm, and local equilibrium instead of enclosure.
Multimodal Workflow
A runnable meeting review workflow that turns audio, screen material, images, and documents into a traceable workbench.
Developer Experience
A local trace viewer that converts Codex and Claude Code JSONL sessions into redacted context and token visualizations.
LLM Eval
A text-evaluation workflow for NIH-style project material, focused on rubric, data scope, badcase review, bias awareness, and human review.
Business Delivery
A real business delivery case covering ERP-style data management, WeChat Mini Program, website structure, and handoff materials.
External Signals
Selected honors that support the industrial AI, sustainability, and engineering-practice parts of the resume.
Field structure
These are the layers I actually operate across when I build. The value is the combination, not any single tool.
Curvature, rhythm, and local equilibrium instead of enclosure.
Capability node
I prefer benchmark-style comparison and traceable decision rules over hand-wavy model claims.
Capability node
I can package technical ability into usable product surfaces instead of stopping at notebooks or scripts.
Capability node
I treat shipping and operating as first-class engineering problems, not afterthoughts.
Capability node
My differentiator is not only building apps, but designing personal and team systems that make intelligence reusable.
Measured history
For readers who want the conventional signal: education, awards, and the milestones that shaped my operating profile.
Curvature, rhythm, and local equilibrium instead of enclosure.
Education
M.S. in Applied Data Science and Statistics
B.A. in Mathematics and Computer Science
Awards
Recognized for drone-assisted power plant inspection work.
Academic recognition during undergraduate study.
Milestones
Participating in real-time warning system design for EV shipping safety.
Completed reproducible modeling and method comparison on public health data.
Worked on high-risk industrial inspection transition from manual to drone-based workflow.