Resume
Software Engineer
I build production AI agent systems — from memory architecture and context engineering to autonomous agent loops and real-time voice. 10+ products shipped end-to-end as sole engineer. My daily workflow is 100% AI-driven.
Summary
Software engineer with 10+ years of experience shipping production systems. Currently building Polaris — a personal AI agent with persistent memory, context engineering across limited attention windows, autonomous proactive behaviors, and real-time voice via ElevenLabs Conversational AI.
Built and deployed 10+ products end-to-end as sole engineer — agent systems, payment infrastructure, multi-tenant data architectures, and performance-critical applications. Former Head of Design (Vidy, SF). Berklee-trained in signal processing and systems design. The rare engineer who obsesses over both the agent orchestration graph and the UI animation curve.
Technical Highlights
Experience
Founding Engineer & Lead Architect
phi.llc ↗Phi Digital • Singapore • 2022–Present
- • Designed and built Polaris — a production AI agent system with persistent memory (semantic embeddings + structured facts), context engineering that assembles life state from 9+ data sources into bounded context windows, and 6 autonomous proactive behaviors (birthday reminders, relationship fade detection, spending anomalies, streak encouragement, morning briefings, mood monitoring)
- • Engineered agent memory architecture: dual-storage learning (embedding-backed semantic memory + key-value learned facts), real-time fact extraction from every conversation, and daily insight summarization that detects behavioral patterns across 24-hour windows
- • Built context engineering pipeline that constructs optimal agent context within limited attention — queries calendar, relationships, goals, expenses, trackers, and mood data in parallel, formats into structured prompt sections, and ranks by relevance
- • Implemented agent tool calling with 15+ client-side tools: event creation, expense tracking, contact management, pending reply resolution, social handle storage — each with verification and error recovery
- • Integrated real-time AI voice via ElevenLabs Conversational AI with Gemini LLM backend — WebRTC streaming, function calling during voice sessions, and interrupt handling
- • Architected multi-tenant data platform with owner_type/owner_id polymorphic ownership across 28 tables, organization support, and row-level security policies — enabling agents to operate within scoped data boundaries
- • Built Life Context API v1 — external platform for third-party agents to read user context and queue actions, with API key auth, scope-based access control, and rate limiting
- • Shipped end-to-end Stripe payment infrastructure: Checkout Sessions, subscription lifecycle webhooks, Customer Portal, and automated plan enforcement
Head of Design
Vidy • San Francisco • 2017–2020
- • Defined technical architecture for the design system serving enterprise ad-tech clients; established component library that reduced iteration cycles by 30%
- • Led cross-functional decision-making across 3 product teams — owned architecture decisions from data model design to frontend implementation strategy
- • Bridged design and engineering: built reusable component primitives, directly contributed to frontend codebase, and mentored junior designers on technical constraints
Software Engineer (Contract)
Multiple Clients • Remote • 3+ years
- • Shipped 10+ production web applications across fintech, music tech, and e-commerce verticals
- • Built multi-chain Web3 wallet system supporting 4 providers across 6 EVM networks with server-side cryptographic signature verification
- • Designed event booking platform processing live Stripe payments with admin workflows, WhatsApp notifications, and automated email pipelines
Featured Projects
Polaris — Personal AI Agent
dayze.com ↗Gemini 2.0 Flash • ElevenLabs Voice • Supabase • pgvector • TypeScript
- • Built autonomous agent runtime — thinks, executes tool calls, verifies results, and recovers from errors across 15+ action types
- • Designed memory architecture with dual storage: semantic embeddings (pgvector, 768-dim) for similarity search + structured learned facts for precise recall
- • Engineered context assembly pipeline that queries 9+ tables in parallel, builds a unified Life State document, and caches with 15-min freshness for optimal context within token limits
- • Implemented 6 proactive agent behaviors running on cron: birthday reminders, relationship fade alerts, spending anomalies, streak encouragement, morning briefings, mood monitoring
- • Integrated real-time voice via ElevenLabs Conversational AI with WebRTC, client-side tool calling during voice sessions, and Gemini as the reasoning backend
- • Built Life Context API for third-party agent access — scoped API keys, rate limiting, action queuing
Dayze — Life Tracking Platform
dayze.com ↗Next.js 14 • Supabase • Stripe • 120+ API Routes • Netlify
- • Built complete SaaS solo — 120+ API endpoints, Stripe subscriptions, admin dashboard, 18K+ notable person profiles
- • Designed multi-tenant organization system with polymorphic data ownership across 28 tables and RLS policies
- • Engineered infinite-scroll virtualized timeline rendering 10,000+ items with cursor-based pagination and prefetch buffering
- • Published 100+ SEO-optimized pages with Schema.org structured data
Coincess — AI Crypto Trading Terminal
GitHub ↗Next.js • Hyperliquid API • WebSocket • Supabase • TypeScript
- • Built real-time trading terminal with live orderbook streaming via WebSocket, automated trading strategies, and direct exchange execution on Hyperliquid
- • Engineered low-latency data pipeline — real-time price feeds, order state management, and cryptographic transaction signing
- • Open source on GitHub — demonstrates production real-time systems and complex state management
ACID-303 Synthesizer
Live Demo →Web Audio API • Real-time DSP • TypeScript • Canvas
- • Built browser-based synthesizer with real-time DSP — resonant filter modeling, oscillator design, sub-millisecond scheduling
- • Demonstrated math-heavy engineering: filter coefficient computation, FFT-based visualization at 60fps, low-latency audio graph management
Education
Berklee College of Music
B.Mus. Electronic Production & Design — Signal Processing, Algorithmic Systems, Software Engineering
Boston, MA • 2012–2015
Specialized in Digital Signal Processing (DSP), algorithmic composition, and real-time software systems. Coursework in linear algebra, Fourier transforms, filter design, and systems architecture — mathematical foundations directly applicable to embedding spaces, attention mechanisms, and real-time data pipeline design.
Technical Skills
AI / Agent Systems
Agent runtime design, context engineering, memory architecture (semantic + structured), tool calling, autonomous loops, prompt engineering, embeddings, pgvector, RAG
AI Models & Voice
Gemini 2.0 Flash, ElevenLabs Conversational AI, WebRTC voice streaming, function calling, multi-model pipelines
Languages & Frameworks
TypeScript, Python, SQL, Bash, React 18, Next.js 14, Node.js, PHP
Data & Infrastructure
PostgreSQL, Supabase (RLS), pgvector, Prisma, Firebase, cron orchestration, data pipelines, multi-tenant architecture
DevOps & Tooling
CI/CD, GitHub Actions, Docker, Netlify, Vercel, MCP servers, Cursor AI, multi-agent development workflows
Payments & Commerce
Stripe (Checkout, Subscriptions, Webhooks, Portal), billing lifecycle
Edge
- →Production agent system: Polaris — memory, context engineering, tool calling, autonomous loops. Not a prototype, a live product.
- →100% AI-driven workflow: Cursor + multi-agent systems daily. AI is my default way of working.
- →End-to-end ownership: Database schema to production deployment, 10+ products shipped solo
- →Context engineering depth: Built systems that assemble optimal context from 9+ data sources within token limits
- →Design + engineering: Former Head of Design — uncommon product sense in technical execution
Contact
Open to opportunities
Founding / senior engineering roles in AI agents. Singapore, on-site.