Building AI-native Products from 0→1 to Scale
Product & Engineering Leader
I build AI-native workflow products end-to-end — from discovery and architecture to working prototype. Hands-on leadership across multi-agent systems, enterprise copilots, and product delivery at scale.
About
I specialize in bridging the gap between product discovery and pragmatic engineering execution. My approach pairs strong product instincts with the technical depth to lead architecture decisions — from multi-agent AI systems and enterprise copilots to fintech products at scale.
How I Build
Start with Workflow Truth
Map the real process, failure modes, edge cases, and incentives before writing code.
Ship Smallest Slice
MVP with measurable outcomes (time-to-complete, error rate) over feature bloat.
AI-Native = Reliability First
Eval plans, telemetry, guardrails, and HITL gates must exist before scaling.
Scale Through Systems
Quality strategy, delivery cadence, and pragmatic platformization.
Featured Project: AI Sales Agent for Chat-First Marketplaces
A multi-agent LangGraph system for property rentals in Singapore. 4 specialist agents, RAG with 27 gov.sg documents, 12 conversion signals, composable prompts. Demonstrates end-to-end product thinking from discovery through architecture to a working prototype.
Key Product Decisions
- Compound Pain Point Framing — speed alone doesn't convert; the agent must build trust, qualify intent, and surface signals simultaneously
- Trust Acceleration Thesis — 7 trust dimensions: accuracy, compliance, control, tone, lead safety, privacy, escalation reliability
- Tiered Escalation Design — 3-tier strategy replacing binary answer-or-defer: answer directly, answer + flag, or escalate with context
- Info Gap Detection — detect unanswered buyer questions, generate suggested answers, feed seller confirmations back into future conversations
Engineering Highlights
- Multi-agent system: LangGraph supervisor + 4 specialist sub-agents with curated tool subsets
- Composable 6-layer prompt system with personality presets and seller customization
- RAG pipeline: ChromaDB with gov.sg sources, category filtering, citation pipeline
- Hybrid LLM+heuristic signal extraction for 12 conversion signals
- 14 runtime failure handlers, 6 adversarial failure modes tested and patched
Explore the project
Signature Outcomes
- 6-Week MVP to Private Beta: Shipped an AI-native FinOps platform and onboarded 4 pilot SMEs.
- +21% Loan Completion Rate: Improved through funnel diagnosis, UX redesign, and A/B experiments.
- -32% Weekly Incidents: Reduced through structured RCA, refactors, and test strategy.
- -44% Feature Delivery Lead Time: Cut via architecture upgrades and delivery process improvements.
- 1→35+ Team Scale-Up: Built from scratch across engineering, QA, design, and PM.
- 40+ SME & Enterprise Projects: Delivered for 30+ organizations across 15+ industries.
Career Experience
ARiA (FingentiX) — Co-founder, Product & Engineering Lead
May 2025 – Feb 2026 | Singapore
Co-founded an AI-native FinOps platform automating Order-to-Cash for SME finance teams. Multi-agent architecture (LangGraph), evaluation frameworks, and analytics for fast, safe iteration.
Antler — Entrepreneur in Residence
Feb 2025 – May 2025 | Singapore
Selected into Antler's Singapore founder residency. Built first working prototype, led IC pitches, secured pre-seed investment.
AMAY Management — Lead Technical Product Manager
Apr 2023 – Jan 2025 | Singapore
Led product delivery for a B2C digital lending app (300K+ users). +21% conversion, -32% incidents, -44% lead time.
MM-Digital-Solutions — Founder, CEO
Nov 2017 – Dec 2022 | Myanmar
Scaled to 35+, delivered 40+ projects across 15+ industries. Built Digi-Zaay (+27% merchant sales) and Digi-Learn. Raised seed funding.
Tools & Technologies
- Frontend: React (TypeScript), Flutter (Dart), Kotlin, Swift
- Backend: Python (FastAPI), Node.js (Express), PHP (Laravel)
- AI/ML: LangGraph, CrewAI, GPT, Claude, Gemini APIs
- Data & Infra: PostgreSQL, Redis, Pinecone, ChromaDB, GCP, Docker, Terraform
- Dev Tools: Cursor, Copilot, Claude Code, CodeRabbit, CI/CD
Recent Thoughts
- Trust Is the Real Architecture (Mar 23, 2026) — The hardest part of building AI agents isn't the AI. It's everything you build around it so the user actually trusts it with their business.
- AI-Fluent vs AI-Native (Mar 16, 2026) — Most companies calling themselves AI-native are actually AI-fluent. There is a real difference, and it changes what you build.
- Agentic Commerce & Disintermediation (Mar 10, 2026) — The next time a customer buys from a marketplace, they might not even open its app.
- Riding the AI Horse (Mar 3, 2026) — Using LLM models or AI agents for productivity is like riding a horse to make more distance than walking.
- Failure Mode Testing Is Product Design (Feb 23, 2026) — I tested 6 failure modes on my AI agent before a single real user touched it. Every one was predictable.
- Smart Agents Fail at Tool Calls (Feb 18, 2026) — When someone asks "Can our agent do X?", my brain goes to: what tools would it need?