Kimberly (Qijing) Huang
Applied AI & Data Engineer · New York
Graduating December 2026 · open to full-time roles in 2027
I build human-in-the-loop products that turn messy inputs into inspectable decisions—combining multimodal AI, deterministic validation and production-minded interfaces.
Python · TypeScript · SQL · LLM APIs · Next.js · PostgreSQL · MS Applied Analytics at Columbia University
01 / Selected work
Three systems showing the product decisions, technical boundaries and evidence behind the interface—not just the finished screen.
Multimodal operations · 2026
8 deterministic checksReads a photographed construction timesheet with a vision model, then lets eight deterministic rules—not the model’s confidence—decide which rows a person has to look at.
I defined the product boundary and built the authenticated extraction route, the rule engine, the two review stages and the audit trail behind them.
Multi-agent operations · NYC Tech Week
3rd Place OverallTurns one sentence from a student into a grounded work order, shows the eight-step agent trace that produced it, then asks the original reporter whether the repair actually worked.
I co-built the signal-to-verification workflow, the confidence ledger, specialist-agent orchestration and the auditable operator experience.
AI learning product · 2026
Deployed · 3 mini-labsInteractive STEM mini-labs that classify what the learner actually misunderstands, then reply with the smallest useful Socratic question instead of the solution.
I designed and built the misconception-first learning flow, the interactive experiments, the coaching route and the teacher-ready session report.
Additional builds
Browser productivity
A Chrome MV3 new-tab workspace that groups open tabs locally and only calls a model when you ask it to name the missions.
No backend · 12 featuresSoftware supply-chain intelligence
Reads a dependency manifest, crawls live web signals for seven categories of supply-chain risk, and returns a scored report with copy-paste repair commands.
Pipeline every 15 min02 / Experience
Work spanning enterprise AI delivery, product analytics and evidence-based decisions.
2026—Now
Deutsche Windtechnik · Houston / Global
Evaluate and build AI-assisted workflows for enterprise operations, translating ambiguous stakeholder needs into bounded products with explicit privacy, validation and human-review controls.
2024—25
Temu · Shenzhen
Analyzed acquisition, activation, retention and monetization across more than 20 markets; automated recurring reporting and translated funnel behavior into growth decisions.
2023
Securities research and corporate finance
Built an early foundation in rigorous analysis, financial reporting and executive communication.
03 / About
I am most useful where the requirements are moving and a credible product needs both engineering discipline and a clear human boundary.
The hard part is the boundary: what the model may decide, what code must decide, and what a person has to sign.
So I separate probabilistic perception from deterministic checks, keep the evidence behind a recommendation visible, and design the point where automation stops before it costs someone money or trust.
MS Applied Analytics · GPA 4.0 · Expected December 2026
BSc Applied Finance · Top 1% · ACCA certified
Multimodal extraction, agent orchestration, evaluation, retrieval, guardrails and human review.
Python, TypeScript, SQL, Next.js, FastAPI, APIs, GitHub Actions and Vercel.
Analytics engineering, experimentation, funnel analysis, financial analysis and executive communication.
