Skip to main content

Founding Engineer·Product Engineer·Frontend Engineer·LLM Agents·Memory & Personality Engines·RAG·Full-Stack AI

MonicaWu

I build AI agents that feel human — memory, personality, and the interface, owned end to end. Ex-Microsoft, IQVIA.

10 yrsProduction software shipped
2 in 2026Production AI systems shipped
100% recallHITL AI, seeded ground truth
  • Large Language Models (LLM)
  • RAG
  • Agentic Workflows
  • Vector Databases
  • LLMOps
  • Python
  • TypeScript
  • React / Next.js
  • Angular
  • Claude Code
01

About

I build AI agents that feel human. I own end to end: memory architecture, personality scaffolding, and user-facing interfaces.

I'm a founding-engineer-type product engineer with a frontend backbone: 10 years shipping production software end to end, and 2 production AI systems shipped in 2026, each with per-stage evals and production observability. I build memory-first architectures — semantic chunking, vector retrieval, per-stage eval harnesses, real-time intent routing — in Python, TypeScript, React, and Angular.

Digital Twin — an agentic RAG agent with structured tool-calling and real-time intent routing (Python, OpenAI, ChromaDB), live on Hugging Face Spaces since May 2026. Reconciliation Workbench — human-in-the-loop AI exception review for multi-jurisdiction tax compliance: 100% recall, zero false positives against seeded ground truth, with a full audit trail.

Software that moves $200M a month doesn't get to break — that's where I learned to ship. Before AI: an enterprise payments platform processing $200M+/month (Accenture), the design system behind healthcare apps used by physicians and pharmacies nationwide (Inovalon), WCAG-accessible components deployed across ~50,000 Microsoft partner websites, and clinical research apps across 9 client sites (IQVIA). When I migrated that component library across 6 major Angular versions, downstream teams saw zero regressions. That's the bar I build to.

I started in design — a BA in Art & Design — then Code Fellows, then a decade of production TypeScript, React, Angular, and Node.js. I own the product from design system to model call, and I treat AI uncertainty as a UX problem: human-in-the-loop review, guardrails, intent routing. I practice AI-assisted development daily (Cursor, Claude Code), cutting feature delivery time by ~40% while keeping every change review-ready.

That AI work started at IQVIA, where our clinical apps had to run locally on-site and client hardware varied site to site — so I tested whether LLM inference could clear that bar: Llama 3.1 8B on Ollama on constrained hardware as a worst-case baseline. Self-initiated, no mandate. It established the AI roadmap for regulated clinical deployments.

The conversations I want: founding engineer roles at seed-stage AI companies building consumer LLM products, and senior, remote product or frontend engineering roles shipping LLM features to real users — where the interface and the model behind it are the same job. Remote (SF / NYC / SEA), distributed-first.

10 yrsShipping production software end to end
2Production AI systems shipped in 2026, each with per-stage evals and observability
$200M+/moPayments platform I shipped transaction-critical features for — checkout abandonment down ~15%
~40%Faster feature delivery with AI-assisted development (Cursor, Claude Code)
wumonica.eng@gmail.com
02

Projects

Live · Auth-gated

Reconciliation Workbench — human-in-the-loop AI for tax compliance

Rules engine + LLM triage + HITL review + golden-dataset evals · multi-jurisdiction (US sales tax, EU CTC e-invoicing) · Aug 2026 — present

Human-in-the-loop AI exception review for multi-jurisdiction tax compliance. Deterministic core, AI at the edges, humans at the points of consequence: a rules engine detects six exception classes, an LLM triage layer explains each exception and proposes a disposition, a reviewer approves or overrides, and every decision lands in an exportable audit trail. An eval panel scores every run against seeded ground truth — 100% recall, zero false positives.

Measured on every run

Golden-dataset eval panel measures precision and recall against seeded ground truth on each run: 100% recall, 1.0 precision, zero false positives.

Jurisdictions as configuration

Adding a country is one JSON entry and zero downstream code changes. A declarative connector layer normalizes native-format feeds (semicolon CSVs, comma decimals, nested-JSON e-invoice statuses) into one canonical model.

Six exception classes

Rate, arithmetic, duplicate invoice, e-invoice linkage, orphan e-report, and return tie-out — detected deterministically, explained by the LLM, decided by a human.

Audit trail & governance

Every approve/override is logged to an exportable audit trail. All data synthetic; deployed auth-gated on Hugging Face Spaces; security-scanned with Bandit and pip-audit.

  • Python
  • Gradio
  • Human-in-the-loop (HITL)
  • Rules engine
  • LLM triage
  • Golden-dataset evals
  • AI governance
  • Synthetic data
  • Hugging Face Spaces
Live · Production

AI Digital Twin — production agentic RAG agent

Agentic RAG + LLM · structured tool-calling · real-time intent routing · per-stage eval harness · live on Hugging Face Spaces · Apr 2026 — present

Production agentic RAG agent with structured tool-calling and real-time intent routing that answers natural-language questions about my professional background in my own voice — GPT-4.1 Mini grounded by semantic search over ChromaDB. Shipped to production in 6 weeks; live and maintained independently since May 2026.

Semantic-chunking redesign

Semantic-chunking redesign improved retrieval coherence over the naive baseline — retrieval returns whole, word-aligned passages instead of mid-word fragments.

Per-stage eval harness

Chunking, embedding, retrieval, context assembly, and generation each have their own evals and can be swapped independently — when quality degrades, the harness shows which stage did it.

Real-time agentic routing

Real-time agentic routing delivers hire/collaboration intent to me within seconds of a visitor interaction.

Production guardrails

Privacy and integrity guardrails: no personal contact disclosure, no fabricated opinions, no binding commitments. Latency, cost, and errors monitored in production.

How it's evaluated

Retrieval quality is verified, not assumed: before/after comparison across the chunking change on a fixed question set, guardrail behavior spot-checked against adversarial prompts, and semantic cluster separation inspected in the companion pipeline to confirm passages group coherently rather than fragmenting.

  • Python
  • OpenAI API
  • GPT-4.1 Mini
  • ChromaDB
  • Gradio
  • Hugging Face Spaces
  • Agentic RAG
  • Tool-calling
  • Eval harness
  • Guardrails
  • LLM observability
Pipeline · Research

Reproducible RAG Pipeline — semantic chunking, embeddings & cluster visualization

Open-source Jupyter walkthrough · Apr 2026 — present

A hands-on Jupyter notebook that walks through the core building blocks of a Retrieval-Augmented Generation (RAG) pipeline — from raw document ingestion through intelligent chunking, embedding generation, semantic clustering, and interactive 3D visualization. The demo document is the Netflix Culture Memo (June 2024). The component engineering behind my production agent.

  • Python
  • ChromaDB
  • Embeddings
  • Semantic chunking
  • UMAP
  • KMeans
  • Cluster eval
  • Vector search
  • Data visualization
  • Jupyter
Portfolio · Design + build

wumonica.com — portfolio

Designed and built end to end in Next.js, TypeScript, and Tailwind CSS · typography, visual hierarchy, component system, accessibility, performance owned as one surface · 2016 — present

This site. A hand-built design system — Cormorant Garamond + DM Mono, dark and light themes, semantic HTML — rendered by Next.js (App Router, static export) and React with Tailwind CSS v4 utilities; WCAG 2.2 AA as an acceptance criterion, zero third-party requests, self-hosted fonts, and a hardened CSP. Showcases two production AI systems and 10 years of high-stakes UI.

  • Next.js
  • React
  • TypeScript
  • Tailwind CSS
  • Design systems
  • Visual hierarchy
  • Accessibility / WCAG 2.2 AA
  • Performance
03

Stack

Top skills

  • Large Language Models (LLM)
  • Retrieval-Augmented Generation (RAG)
  • Agentic Workflows
  • Vector Databases
  • Large Language Model Operations (LLMOps)

LLM agents & memory

  • LLM Agents
  • Memory Architecture
  • Personality Scaffolding
  • Agentic AI Development
  • Tool Calling / Function Calling
  • Agent Orchestration
  • Structured Outputs
  • Intent Routing
  • OpenAI API
  • GPT-4.1 / Mini
  • Ollama · Llama 3.1
  • Local LLM Inference

Retrieval & context

  • Retrieval-Augmented Generation (RAG)
  • ChromaDB
  • Vector Databases
  • Embeddings
  • Semantic Search
  • Semantic Chunking
  • Prompt Engineering
  • Context Engineering
  • Context Window Management

Evals, guardrails & HITL

  • LLM Evaluation
  • Per-stage eval harnesses
  • Golden-dataset evals
  • AI Guardrails
  • Human-in-the-Loop (HITL)
  • AI Governance
  • LLM Observability
  • Rules Engines
  • Synthetic Data

Ship & deploy

  • Production AI Systems
  • Hugging Face Spaces
  • Gradio
  • CI/CD
  • Production Deployment
  • Test Automation
  • Bandit / pip-audit
  • 0→1 Product Build

Languages & frameworks

  • Python
  • TypeScript
  • JavaScript
  • React
  • Next.js
  • Tailwind CSS
  • Node.js
  • Angular · RxJS
  • Redux
  • REST APIs
  • MongoDB
  • Pandas
  • Jupyter

AI-assisted engineering

  • Claude Code
  • Anthropic Claude
  • Claude Cowork
  • Cursor

Product & design

  • Design Systems
  • Component Libraries
  • Accessibility / WCAG 2.1–2.2 AA
  • Interaction Design
  • Visual Hierarchy · Typography
  • UX
  • Data Visualization
  • i18n
  • Performance Tuning

Domains

  • Consumer LLM Products
  • FinTech · Payments ($200M+/mo)
  • Tax & Regulatory Compliance
  • Healthcare IT · Clinical research
  • Media (NBCNews.com)
  • Regulated / high-reliability
04

Experience

Aug 2025 — present · 1 yr 1 moWumonica Studio (Independent) · Self-employed · Remote

Senior Frontend Engineer — Agentic RAG & Human-in-the-Loop AI

Shipped 2 production AI systems in 2026 with full-stack ownership: Digital Twin (agentic RAG with Python, OpenAI API, ChromaDB; live since May 2026) and Reconciliation Workbench (human-in-the-loop AI with Gradio, Python, and audit trails; 100% recall, 1.0 precision against seeded ground truth). Built memory-first architectures: semantic chunking, vector retrieval, per-stage eval harnesses that isolate the degraded stage, and real-time intent routing — with latency, cost, and errors monitored in production. Made jurisdictions (US sales tax, EU e-invoicing/CTC) pure configuration via a declarative connector layer: a new country is one JSON entry, zero code changes. Practice AI-assisted development daily (Cursor, Claude Code), cutting feature delivery time by ~40% while maintaining high code quality. Designed and built this portfolio site end to end in Next.js, TypeScript, and Tailwind CSS — typography, visual hierarchy, component system, accessibility, and performance as one integrated surface.

  • LLM Agents
  • Agentic RAG
  • Human-in-the-loop
  • Evals
  • Guardrails
  • Python
  • OpenAI API
  • ChromaDB
  • Next.js
  • Claude Code
  • Cursor
Sep 2023 — Aug 2025 · 2 yrsInteliquet, an IQVIA business · Full-time · Remote

Software Development Engineer 4

Self-initiated a local LLM inference study (Ollama, Llama 3.1) on constrained hardware that established the AI roadmap for regulated clinical deployments. Owned frontend features end to end for clinical research apps across 9 client sites: ran requirements discovery with stakeholders, shipped production TypeScript/Angular components, and resolved post-launch issues within 24 hours. Consolidated 10+ Angular components into a shared library and shipped i18n across 2 production apps, reducing duplicate work across 3 product teams by ~25%. Accelerated delivery with AI-assisted development (Cursor) in a regulated healthcare codebase, keeping every change review-ready and audit-compliant.

  • Local LLM inference
  • Ollama · Llama 3.1
  • TypeScript
  • Angular
  • Regulated industries
  • i18n
  • Cursor
Jan 2022 — Aug 2023 · 1 yr 8 mosInovalon · Full-time · Remote

Product Engineer — Design Systems

Owned the enterprise design system (NPM component library) end to end, from specs to production — used by healthcare apps serving physicians and pharmacies nationwide. Built reusable, accessible components with typography, interaction design, cross-browser compatibility, and performance as acceptance criteria. Migrated the Angular component library across 6 major versions (v8→v14) with zero regressions, reducing downstream team upgrade effort by ~40%. Authored component specifications adopted as the single reference by internal teams, reducing UI bugs by ~20%.

  • Design systems
  • Component libraries
  • Accessibility
  • Zero-regression migration
  • Angular
  • TypeScript
Jul 2018 — Jan 2022 · 3 yrs 7 mosAccenture · Full-time · Greater Seattle Area · Hybrid

Software Engineer

Delivered transaction-critical frontend features end to end for an enterprise payments platform processing $200M+/month — reduced checkout abandonment by ~15% through UI optimization. Resolved critical production defects through root-cause analysis, restoring payment workflows within SLA 100% of the time. Mentored 2 engineers in advanced Angular (RxJS, state management, lazy loading), accelerating their time-to-productivity by 3 weeks.

  • FinTech · Payments ($200M+/mo)
  • High reliability
  • Root-cause analysis
  • TypeScript
  • Angular · RxJS
  • Mentoring
Jan 2018 — May 2018 · 5 mosMicrosoft · Contract · Redmond, WA · On-site

Software Development Engineer

Delivered a production Angular/TypeScript interface on REST APIs for Microsoft partner and internal teams to publish and manage JSON schemas for automated page generation — used by 10+ partner teams. Collaborated across 2 engineering teams to align API contracts and frontend implementation, shipping on time despite cross-team dependencies.Contract via Design Laboratory Inc.

  • Angular
  • TypeScript
  • REST APIs
  • JSON Schema
Dec 2017 — Feb 2018 · 3 mosWin-Kel · Freelance · Greater Seattle Area · Hybrid

Product Engineer

Architected a 0→1 full-stack product (React, Redux, Node.js, TypeScript) for a 4-person startup — owned product and stack decisions from database schema to UI. Set frontend technical direction and mentored 2 junior engineers through code review.

  • 0→1
  • React
  • Redux
  • Node.js
  • TypeScript
  • Mentoring
Jan 2017 — Apr 2017 · 4 mosMicrosoft · Contract · Greater Seattle Area · On-site

Product Engineer

Built and remediated WCAG-accessible framework components (Angular, React, TypeScript) deployed across ~50,000 Microsoft partner websites. Ensured 100% WCAG 2.1 AA compliance for all components, reducing accessibility-related support tickets by ~30%.Contract via Jetstream Software.

  • Accessibility / WCAG 2.1 AA
  • Angular
  • React
  • TypeScript
Jun 2016 — Oct 2016 · 5 mosNBCUniversal · Contract · Seattle, WA · On-site

Software Engineer

Shipped production fixes, analytics instrumentation, and SEO sitemap improvements for NBCNews.com social and video features serving tens of millions of monthly users. Reduced page load time for video-heavy pages through performance tuning.Contract via Next Step Staffing.

  • JavaScript
  • Performance tuning
  • SEO
  • Analytics
Jan 2016 — Apr 2016 · 4 mosUIEvolution, Inc. · Contract · Greater Seattle Area · On-site

Product Engineer

Built production digital-signage web apps for smart TVs, kiosks, and mobile — with performance and stress testing against real device constraints.Contract through Sixth Ave Studios.

  • React
  • JavaScript
  • Digital signage
  • Stress testing
05

Education & certifications

Issued May 2026

SuperDataScience

AI Engineer Sprint — Capstone: Digital Twin, production RAG agent shipped in 6 weeks

Production-focused AI Engineer Sprint. The capstone became the first version of my Digital Twin agent — shipped to production in 6 weeks and maintained independently since May 2026: agentic RAG, semantic chunking, privacy guardrails, real-time tool-calling, and a per-stage eval harness.

  • AI Agents
  • Agentic AI Development
  • RAG
  • OpenAI API
  • ChromaDB
  • Production deployment
In progress · 2026

Interview Kickstart

Machine Learning Engineering

Machine learning engineering program covering ML fundamentals, deep learning, and model evaluation — deepening the ML foundations behind the production AI systems I ship.

  • Machine Learning
  • Deep Learning
  • Model Evaluation
May 2015 — Oct 2015

Code Fellows

Certificate · Full-stack JavaScript

Immersive, full-time software engineering program focused on production full-stack JavaScript: Node.js and Express REST APIs, MongoDB data modeling, authentication, test-driven development, and deployment. Capstone: conceived, architected, and led a team build of a food-truck locator app over 5 days (OAuth, Google Maps API, JavaScript front end, MongoDB) — personally built the Google Maps integration and the front end. The foundation for a decade of production TypeScript, React, Angular, and Node.js work at Microsoft, Accenture, IQVIA, and Inovalon.

  • React
  • Angular
  • Node.js · Express
  • MongoDB
  • TDD
B.A.

California State University, Los Angeles

Bachelor of Arts (B.A.) · Art — Design Concentration

Concentration in visual systems, typography, hierarchy, and interaction design: the foundation for the design systems and component specifications adopted as the single reference at Inovalon.

  • Interaction design
  • Design systems
  • Typography
  • Visual communication
06

Get in touch

If your AI feature works in the demo but not in production, message me and I'll tell you how I'd approach it.

The conversations I want: founding engineer roles at seed-stage AI companies building consumer LLM products, and senior, remote product or frontend engineering roles shipping LLM features to real users — where the interface and the model behind it are the same job. Remote (SF / NYC / SEA), distributed-first.