Open to new work opportunities

Building Vision AI pipelines for real-world scenarios.

Machine Learning Engineer with 5+ years building end-to-end Computer Vision and Generative AI systems with fine-tuned LLMs and vision models that turn visual detections into meaningful insights. Pick an industry below to explore Amiir's know-how.

Smart City & Automotive

Multi-plate ANPR with blacklist routing, and plant safety-zone compliance.

Explore Sandbox

Oil & Gas / HSE

LLM housekeeping assessment and strict PPE compliance pipelines.

Explore Sandbox

Infrastructure

Road defect & asset inspection, LLM false-positive filtering, drone patrol.

Explore Sandbox

Smart City & Automotive

AI deployments for urban analytics and plant HSE.

ANPR models deployed on city CCTV around Klang Valley — the live frame below is an example CM045F on Lebuhraya Sultan Iskandar. A scan runs the real pipeline order: detect every vehicle first, then localise and read the plate inside each box. Every read is cross-referenced against the blacklist, and a confirmed hit goes to the re-identification service, which reconstructs the vehicle's route across the camera mesh.

Detects every vehicle in the frame, then reads the plates off them.

AWAITING SCAN — CM045F / LEBUHRAYA SULTAN ISKANDAR
Inference: —

Vision AI monitoring inside major plants in Malaysia for workers safety. Use cases include operator detection in hazardous area. A LLM layer reads the occupancy against the rule and writes the ticket.

Qwen2-VL · Violation Reasoning NO ACTION

HAZARD_ZONE_A
PERSON_01 · IN_ZONE
PERSON_02 · IN_ZONE
PERSON_03 · WALKWAY
IN_ZONE 2 / MAX 1

Oil & Gas / HSE

Housekeeping assessment and high-precision PPE compliance.

PATH_CLEAR
CHAIR_STOWED
PATH_CLEAR
CHAIR_STOWED
CHAIR_STOWED
LIQUID_SPILL
BOTTLES_ON_FLOOR
BIN_OVERFLOW
FLOOR_LITTER
CHEMICAL_SPILL
TRAILING_CABLE
BROKEN_GLASS
FOOD_AND_DRINK
CLEAN LAB
CLUTTERED LAB

Have been deployed at one of Petronas campus to monitor lab housekeeping. Drag the divider to set how much of the frame is in a cluttered state while an LLM reasoning layer (Qwen) reads the detected objects and returns a 5S condition assessment.

50% cluttered
LLaMA · Condition Assessment —
— 5S SCORE / 100

NO_HARD_HAT
NO_HARD_HAT · NO_HI_VIS
PPE: 2 OF 6 NON-COMPLIANT

One frame, six people, one verdict per person. Detection runs per body — hard hat, FR hi-vis, then a compliance rule on top — so a mixed crew doesn't collapse into a single site-wide pass/fail.

2 violations / 6 persons CAM-7 SOUTH YARD

Infrastructure

Road asset inspection and remote aerial patrols.

Dashcam and drone road surveys deployed in Malaysia, Singapore and Japan. Two tier system to ensure in reduction of false positive cases.

Survey Output — 1 frame · 61 km/h
POTHOLE 0.94
POTHOLE 0.86
ALLIGATOR_CRACK 0.91
ALLIGATOR_CRACK 0.84
ALLIGATOR_CRACK 0.84
SIGN_DAMAGED 0.90
SIGN_DAMAGED 0.70
DET: 6/6 boxes

Detection of common road defects and broken road assets. Addition of a VLM verification pass that re-reads each crop in context and does one of three things with it: confirm it, drop it, or keep the finding and correct its class. Generative AI justify the reasons to flag them as defects.

raw confirmed reclassified suppressed
Precision
61%
False pos. / km
8.2
Verification Log

Verification layer off — all 5 raw detections forwarded to the maintenance backlog.

FORWARDED: 5/5

With autonomous drone patrolling restricted site compound, Vision AI assisted in identifying intruders or suspicious activities.

Re-runs the detector over the same captured frame — this is one real frame from a night sortie, not a state toggle.
Patrol Analysis —

PERIMETER_FENCE
FENCE_CLIMB (0.91)
STATUS: SWEEPING…

About Me

Who is behind the pipelines.

Portrait of Amiir Haamzah
Amiir Haamzah
Machine Learning Engineer · Inference Tech Sdn Bhd
Contact
amiirhaamzah96@gmail.com LinkedIn
Malaysia — open to remote & APAC opportunities

I build computer vision systems for real-world environments, where noisy CCTV feeds, moving dashcams, aerial footage, and challenging industrial conditions are the norm. Over the last 5+ years, I’ve taken projects from data collection and annotation strategy through model training, optimisation, and on-site deployment.

The thread across my recent work is applied generative AI on top of vision: combining detection models with fine-tuned LLMs and VLMs that re-evaluate detections in context before deciding whether to confirm, drop, or correct an alert. The result is a vision pipeline that can reason about context rather than relying on detection alone.

Experience
Vision AI Team Lead - Machine Learning Engineer
Inference Tech Sdn Bhd

Architects Computer Vision deployments end to end and fine-tunes LLMs (Qwen, LLaMA) as reasoning layers on top of them. Leads a data annotation team and owns internal labeling tooling for image, video and segmentation workflows.

Deployment footprint
Malaysia · Singapore · Turkmenistan · Japan

Smart-city ANPR and landslide/dumping detection (~50 cameras, MPS), Perodua Smart Mobility Plant safety compliance, Petronas Research housekeeping monitoring (179 cameras), international PPE deployment for PCTSB, national road surveys and drone patrols.

Education
MSc, Manufacturing & Mechatronic Engineering
UMPSA — Research Mode · 2021– August 2026

Completed while working full-time as a Machine Learning Engineer.

BEng (Hons), Mechatronics Engineering
Universiti Malaysia Pahang Al-Sultan Abdullah · 2017–2021
Diploma, Engineering Technology (Marine)
Universiti Kuala Lumpur · 2014–2016
Stack
PyTorch YOLOv8 / RT-DETR Segmentation TensorRT / ONNX Generative AI LLM fine-tuning Qwen-VL / LLaMA RAG OpenCV Docker Annotation Ops
Publications
  1. Keras Implementation in Detecting Intracranial Hemorrhage and Multiclass Classification of Subtypes via Transfer Learning and Classifiers Selection
    Mekatronika — Journal of Intelligent Manufacturing & Mechatronics · 2024
    doi.org/10.15282/mekatronika.v6i2.11358
  2. The Diagnosis of COVID-19 by Means of Transfer Learning through X-ray Images
    ICCAS — IEEE International Conference on Control, Automation & Systems · 2021
    doi.org/10.23919/ICCAS52745.2021.9649899
  3. The Diagnosis of COVID-19 Through X-Ray Images via Transfer Learning and Fine-Tuned Dense Layer on Pipeline
    Mekatronika — Journal of Intelligent Manufacturing & Mechatronics · 2021
    doi.org/10.15282/mekatronika.v3i2.7161
  4. The Diagnosis of COVID-19 Through X-Ray Images via Transfer Learning Pipeline
    Springer — Lecture Notes in Electrical Engineering · 2021
    doi.org/10.1007/978-3-030-70917-4_36