PSK

Multimodal AI Technical Lead at GovTech Singapore

Singapore

Pye Sone Kyaw

I build vision and multimodal AI systems that ship to production — not papers, not demos. From discovery with stakeholders to deployed solutions across 12+ government agencies.

Multimodal AIVision AIFull LifecycleProduction Systems
About

Background

Multimodal AI Technical Lead at GovTech Singapore, bridging R&D and production. I own the full AI solution lifecycle: discovery with agency stakeholders, rapid POCs, iteration with users, and deployment to production.

I lead a team delivering vision and multimodal AI capabilities across 12+ government agencies — from deepfake detection for national elections to heritage artifact cataloguing to biodiversity recognition. I also architect internal platforms (MM Hub, Vision Studio) that help other teams go from prototype to production faster.

I ship in weeks using AI-forward development practices.

Experience

Work History

Education

Academic Background

Skills

Technologies

I work across the full stack — from model training to edge deployment to production APIs. I embrace AI-forward development: Claude Code, agentic workflows, and rapid iteration.

PyTorch
Transformers
Accelerate
Diffusers
vLLM
Ollama
LanceDB
Docker
TensorRT
Jetson
React
Next.js
TypeScript
Node.js
FastAPI
Claude Code
Cursor
MCP
Projects

Selected Work

A collection of projects I've built, from research prototypes to production systems.

Vision Studio

Vision Studio

Playground and POC deployment platform for the Whole-of-Government AI Platform. Where validated vision and multimodal AI capabilities go live for government agencies.

React
Next.js
Tailwind
+2
Multimodal AI Hub

Multimodal AI Hub

Internal experimental prototyping platform for multimodal AI. Rapid capability development sandbox where teams explore VLMs, multimodal RAG, and edge-cloud hybrid solutions before production.

React
Next.js
Tailwind
+2
ScopeSite

ScopeSite

Construction progress monitoring for a public housing agency. Computer vision + photogrammetry to track site changes from drone imagery.

Computer Vision
Photogrammetry
Drones
+1
Biome.ai

Biome.ai

Species recognition for a national parks agency. Vision foundation models + multimodal RAG for biodiversity monitoring in the field.

Multimodal AI
CLIP
VLMs
+2
Relic.ai

Relic.ai

Artifact cataloguing for a national heritage institution. Multimodal AI to search, classify, and understand museum collections — reduced cataloguing time by 74%.

Multimodal AI
CLIP
VLMs
+6
Balefire

Balefire

Illegal smoking detection for an environmental enforcement agency. Edge AI on Jetson Xavier — reduced false alerts by 80%, enabling real-time field operations.

Edge AI
Jetson Xavier
TensorRT
+4
MDP Playbook

MDP Playbook

Open-sourced robotics project for NTU's notoriously difficult MDP course. Now used by 50%+ of students as their starting base.

Python
YOLO
Some heuristics
+4
RecycleTree

RecycleTree

Hackathon project: AI-powered recycling classification. Won multiple awards across different competitions.

React
Tailwind
Supabase
+4
Contact

Let's Connect

Open to Forward Deployed Engineer, Applied AI, and Solutions Engineer roles. Also happy to chat about multimodal AI, rapid prototyping, or shipping AI in government.