hazwan.
ML research engineer · Kuala Lumpur

I build
models that
ship.

Nine years training, compressing and deploying deep learning systems across speech, vision and multi-agent LLM pipelines, from synthetic data to TensorRT on a Jetson to a weekly production run nobody has to babysit.

A sculpted black glass and metal sound wave that breaks apart into small cubes, a few glowing orange
9+ yrs
training & shipping ML
10+
models trained & deployed
264 h
code-switched speech in prod
12×
parallel LLM QA agents / episode
01Selected work

Things I've built
and shipped.

Case study · 2025 — now · live weekly

An autonomous multi-agent subtitle pipeline

A podcast episode drops, and a manager agent fans the transcript out to parallel LLM workers, gates every edit with deterministic scripts, and ships bilingual Malay + English subtitles. It writes its own retrospective when it's done.

Read the architecture →
Two black glass masks, one with a glowing orange crack
MIMOS · national R&D2026

Multimodal deepfake detection

Matched real/fake evaluation protocol (AUC / FPR on honest negatives), a 13-paper audit narrowed to 5 audio-visual models, and fine-tuning on Malaysian demographic data.

audio-visualeval designfine-tuning
A black glass hub connected to metal cubes, one glowing orange
Nervesis2025 — 26

Agentic enterprise analytics

LangGraph multi-agent system with plan decomposition and a self-healing data-analyst agent, plus an LLM-as-judge eval service and a containerised MCP server for live financial data.

LangGraphRAGMCPLLM-as-judge
A wireframe crate turning into a solid glass crate
Ascent Solutions · SG2021 — 24

Synthetic data & sim-to-real

Photorealistic Unity/Blender data pipeline with domain randomisation for rare edge cases, QAT + TensorRT for edge budgets, and multi-camera detections projected into a live warehouse digital twin.

UnityBlenderQATTensorRT
A microchip wearing a tiny orange hard hat
Inference Tech Asia2019 — 20

Site-safety vision at the edge

Custom YOLO detectors in an always-on AWS pipeline over live RTMP streams, then ported to Jetson Nano/Xavier with TensorRT + DeepStream for real-time FPS.

YOLOJetsonDeepStream
A metal tile of palm trees seen from above, one marked orange
Aerodyne Group2018

Gigapixel drone inference

Patch-based training and inference over >1B-pixel drone maps that kept global spatial coherence, with a GPS lat/lon for every detected palm tree.

small objectsgeospatial
A stacked tower of blocks topped with a glowing orange cube
Grab AI for SEA2019

Top submission, computer vision track

Regional challenge entry focused on spatial reasoning. Also top 5 at the Seedance “Build with AI” hackathon 2026 (AI video generation).

competitionCV
Training & fine-tuningArchitecture searchQAT · pruning · TensorRTSynthetic dataSpeech & ASR evalDeepfake detectionMulti-agent orchestrationLLM-as-judge Training & fine-tuningArchitecture searchQAT · pruning · TensorRTSynthetic dataSpeech & ASR evalDeepfake detectionMulti-agent orchestrationLLM-as-judge
02Experience

Nine years,
cloud to edge.

BEng, Cardiff University

03Contact

Let's build
something.

Research roles, consulting, or a model that has to work outside the notebook. Send a note and I'll reply by email.