About Experience Skills Highlights Side Project Contact

Diem (Paige) Chau

AI Product Manager · Conversational AI & Enterprise NLU

I help enterprises ship conversational AI that works in regulated, high-stakes environments. At JPMorgan Chase, I own the chatbot and NLU channel for Apple Card customers — translating complex AI systems into products people actually use.

See My Work → Get in Touch
8+
Years in tech & AI/ML
$1M+
Cost savings delivered
0+
Production intents scoped
About

Shipping AI that works
at enterprise scale.

I'm a product leader focused on conversational AI, LLM applications, and intelligent automation — shipping systems that hold up under compliance scrutiny and real-world edge cases.

At JPMorgan Chase, I own the chatbot and NLU channel for Apple Card customers — a cross-company integration with Apple's system, built on Amazon Connect, Amazon Lex, and Lambda. I govern 250+ intents across payments, lending, and disputes, defining the API contracts and escalation paths behind each one.

Before JPMC, I spent four years at FPT USA Corp. deploying AI for Fortune 500 clients — LLM-powered enterprise search, computer vision defect detection, and RPA automation that generated $1M+ in documented cost savings. I bring hands-on delivery experience and PM rigor to every AI initiative I touch.

On the side, I build to keep my hands dirty — most recently Axiom, an AI claims-intelligence platform with a live Claude-powered demo you can try below.

Current Role
PM — Chatbot & Conversational AI
JPMorgan Chase · Consumer Banking · Plano, TX
Focus Areas
NLU · LLM · RPA · Computer Vision
Regulated Industries · Enterprise AI
Background
Product + AI Delivery
8+ years across tech & AI/ML
Experience

Where I've been.

Oct 2024 – Present
JPMorgan Chase
Current
JPMorgan Chase · Consumer & Community Banking · Plano, TX
Product Manager — Chatbot & Conversational AI
Own the chatbot architecture on Amazon Connect for Apple Card customers — a cross-company NLU integration with Apple's Arsenal system covering 250+ intents across payments, lending, and disputes. Define the full intent matrix, API contracts, and escalation paths. Drive fraud detection strategy for demographics-change events while maintaining compliance guardrails. Align 5+ internal teams and Apple on API specs and handoff design.
Amazon Connect Amazon Lex NLU Intent Design Lambda Cross-Company Integration
Aug 2020 – Oct 2024
FPT USA Corp.
FPT USA Corp. · AI Center · Richardson, TX
Senior Technical BA → AI Product Lead
Shipped LLM-powered enterprise search with 40% retrieval time reduction, adopted daily org-wide. Deployed computer vision defect detection (YOLO, OpenCV) with 35% QC time savings. Led ML predictive maintenance reducing downtime by 20%. Delivered $1M+ in cost savings across 5 AI/ML and RPA projects for Fortune 500 clients in oil & gas, insurance, and manufacturing.
LLM · RAG Computer Vision YOLO · OpenCV RPA Amazon Bedrock Azure AI
Oct 2019 – Aug 2020
BAP Software
BAP Software · Ho Chi Minh City, Vietnam
Product Manager
Pitched and launched an AI-driven smart fashion app — led 8-person team from concept to beta, securing $1M pre-seed investment. Improved team delivery velocity by 20% introducing Agile/Scrum ceremonies.
AI Product Agile / Scrum Pre-seed
2017 – 2019
Actility
Actility · Paris, France
Digital Strategy Executive
2014 – 2017
Galaxy Play
Galaxy Play · Ho Chi Minh City, Vietnam
Digital Strategy Executive
Skills

What I bring to the table.

A rare blend of hands-on AI delivery, PM rigor, and technical depth — built across regulated enterprise environments.

🗣️
Natural Language Understanding
Intent taxonomy design, entity extraction, confidence thresholds, fallback & escalation logic. Scoped 250+ production intents at JPMC. Amazon Lex, Arsenal NLU handoff design.
🤖
LLM & RAG
LLM-powered enterprise search, RAG architectures, Amazon Bedrock, Azure AI. Chunking strategy, relevance evaluation frameworks, and prompt engineering at production scale.
👁️
Computer Vision
YOLO v8, OpenCV, real-time defect detection pipelines for manufacturing QC. Deployed for Fortune 500 clients — 35% QC inspection time reduction.
⚙️
RPA & Automation
End-to-end process automation, Automation Anywhere. $1M+ in documented cost savings across oil & gas, insurance, and manufacturing Fortune 500 clients.
💻
Vibe Coding
Building and shipping AI-powered prototypes with code — Python, prompt engineering, rapid iteration with Claude, Cursor, and modern AI dev tools. This portfolio is a product.
🤝
Stakeholder Alignment
Bridging engineering, legal, compliance, and business. Cross-company alignment with Apple and 5+ JPMC internal teams — comfortable operating where the stakes are real.
Highlights

Impacts I've driven.

High-impact AI initiatives across conversational AI, computer vision, and automation — delivered in regulated, enterprise environments.

🤖
Apple Card Chatbot — Cross-Company NLU Integration
JPMorgan Chase · Consumer Banking
Ongoing
Own the end-to-end chatbot architecture on Amazon Connect for Apple Card customers. A cross-company NLU integration with Apple's Arsenal system — contact flows receive structured intent payloads from Apple's NLU, invoke Lambda for routing logic, and use Amazon Lex for in-session handling. I govern the full intent matrix across payments, lending, and disputes, with defined API contracts and escalation paths for each.
🎯 250+ intents in production · Live for all Apple Card customers
🔍
Delivered
Enterprise LLM Search
FPT USA Corp.
LLM-powered enterprise document search on RAG architecture. Reduced retrieval time by 40%, adopted daily org-wide. Built on Amazon Bedrock with custom chunking and relevance evaluation.
📄 40% reduction in document retrieval time
👁️
Delivered
CV Defect Detection — Manufacturing QC
FPT USA Corp.
Real-time computer vision pipeline using YOLO v8 and OpenCV for manufacturing defect detection. Deployed for Fortune 500 manufacturing clients across production lines.
📉 35% QC inspection time reduction
⚙️
Delivered
ML Predictive Maintenance
FPT USA Corp.
ML-based predictive maintenance for manufacturing clients — sensor data ingestion, anomaly detection, alert thresholds, and integration with maintenance ticketing systems.
⚙️ 20% reduction in operational downtime
🔄
Delivered
RPA Automation Portfolio
FPT USA Corp.
End-to-end process automation across 5 AI/ML and RPA projects for Fortune 500 clients in oil & gas, insurance, and manufacturing. Automation Anywhere implementation and process redesign.
💰 $1M+ in documented cost savings
Side Projects

AI tools I'm building.

Experiments at the intersection of AI and product — building to learn, ship, and demonstrate what good AI product thinking looks like.

🏥
Axiom — AI Claims Intelligence Platform
Meridian Health · AI Product Manager, Claims Intelligence · Capstone Project
Live AI Demo ↓
Meridian Health's 1.2M members faced a claims backlog averaging 12 days to resolution, a 23% adjuster error rate, and thousands of members navigating denials without support. I led product strategy for Axiom — an AI claims automation platform that auto-approves low-complexity claims in seconds, routes edge cases to adjusters with AI-generated summaries, and gives members plain-language explanations of every decision.
Key PM Decision
Tiered Autonomy
Auto-approve Tier 1 (<$500, in-network, clean codes). Human gate on Tier 2–3. Reduced adjuster load 61% without sacrificing accuracy on complex cases.
Key PM Decision
Explainability First
Every decision — approve or deny — generates a plain-language rationale citing policy section, ICD-10 code, and coverage rule. HIPAA-compliant audit trail on every output.
Key PM Decision
Member-Facing Layer
Added conversational intake channel so members submit claims in natural language — no portal login, no forms. Reduced call center volume 34% in pilot.
⚡ Avg. resolution: 12 days → 4 min (Tier 1)
🎯 Adjuster error rate: 23% → 6%
📉 Call center volume ↓ 34%
✅ 61% of claims auto-approved
How it's built — the technology
🧠
LLM Reasoning Core
Claude (Anthropic) · claude-sonnet-4-6
Claude handles the unstructured-to-structured work: parsing a free-text claim into provider, service, CPT/ICD-10 codes, and amount, then reasoning over plan rules to produce a triage tier, a decision, and a plain-language rationale. Chosen for strong instruction-following and reliable structured output under a strict system prompt.
🗂️
Knowledge & Rules Layer
Plan policy · ICD-10 / CPT · Eligibility
Plan benefits, deductibles, coinsurance, and coverage exclusions are encoded as ground-truth context (in production: a retrieval layer over policy documents + a claims/eligibility database). This keeps the model grounded in real coverage rules rather than guessing — and makes every citation traceable.
⚖️
Tiered Autonomy Engine
Rules-based routing · Human-in-the-loop
Deterministic guardrails decide how much autonomy the AI gets: Tier 1 auto-approves, Tier 2–3 route to a human adjuster with an AI-generated summary. The model proposes; the policy engine and a human dispose on anything high-stakes. This is the core risk-control design decision.
🛡️
Explainability & Governance
Audit trail · PHI redaction · Citations
Every decision emits a rationale citing the policy section and code it relied on, logged to a HIPAA-compliant audit trail. PHI is redacted before processing. This makes the system defensible to compliance, appealable by members, and reviewable by regulators.
System design
Request flow — one claim, end to end
A member submits a claim in plain language. Axiom extracts structured data, triages complexity, decides, and explains — branching to a human adjuster whenever stakes are high.
Intake
NL · no forms
→
Extraction
Claude
→
Triage
Rules + Claude
→
Decision
Policy engine
→
Explain
Claude
→
Member
+ audit log
↑ At Triage, the claim branches by tier — Tier 1 flows straight through; Tier 2–3 detour to a human adjuster before the decision is finalized.
ChannelsHow members reach Axiom
Conversational intake Member portal Provider submission
AI CoreReasoning & generation
Claude — extraction Claude — triage reasoning Claude — rationale
KnowledgeGround truth
Plan policy & benefits ICD-10 / CPT codes Eligibility & deductibles
OrchestrationRouting & control
Tier-1 auto-approve Tier-2/3 adjuster queue Escalation & appeals
GovernanceCompliance & trust
HIPAA audit trail PHI redaction Decision traceability
Try it live
⚕️
Axiom Claims AI — Interactive Demo
Submit a health insurance claim in plain language. Axiom extracts structured data, triages complexity, and issues a decision with next steps — auto-approving simple claims and flagging complex ones for adjuster review.
Live AI
✅
Tier 1 — Auto-Approve
<$500 · in-network · routine codes
🔍
Tier 2 — Adjuster Review
$500–$5K · out-of-network · specialist
🚨
Tier 3 — Manual + Escalation
>$5K · experimental · fraud flags
Axiom Claims AI
Meridian Health · Claims Intelligence · Powered by Claude AI
Online
Ax
Hi, I'm Axiom, Meridian Health's claims AI. You can submit a claim in plain language — just describe what happened, your provider, and the service received. I'll handle intake, triage, and issue a decision with next steps.

Try one of the example claims below, or describe your own.
⚡ Powered by Claude API · Axiom is a fictional product demo — Meridian Health is simulated · No real PHI or member data used · Not medical or insurance advice
Contact

Let's build something.

Open to senior PM roles in conversational AI, NLU, and enterprise AI products.

diemtnchau@gmail.com → LinkedIn ↗