Malaysia
Hazqeel Afyq
Full Stack AI Engineer
"The key of life is luck coverage maxxing and regret minning."
What I Believe
"Curiosity is the wick in the candle of learning."
"Alone we can do so little; together we can do so much"
— Helen Keller
What I Want to Be
Jack of all trades, master of one. Growth through curiosity and collaboration.
Experience
Full Stack AI Engineer
ACTIVEValiance Health
April 2026 - Current
- ●Architected a 9-cell hierarchical conformal-prediction stack (LightGBM + CatBoost MultiQuantile, cascade anchor, Mondrian CQR at 90% coverage) for healthcare pricing. Added a Nadaraya-Watson kernel for sparse-age-cohort interpolation. Cut MAPE 18%, high-value error 29%, and sparse-cohort bias 60%.
- ●Rebuilt cross-organization p10 / p50 / p90 price lookups using per-org median voting under a hierarchical log1p Bayesian smoother (k = 10). Layered a LightGBM + CatBoost procedure ensemble: raised pinball_p50 45%, halved cross-seed variance, and cut n = 1 procedure error from 119% to 70%.
- ●Shipped daily revenue forecasting via a two-track pipeline (5-model linear / non-linear ensemble vs Optuna-tuned XGBoost) with conformal 90% PI. Cut MAE 40%-71% at R-squared 0.66-0.79; consolidated sparse per-item forecasts into cluster-pooled models for 97% count reduction and ~10x holdout accuracy.
- ●Diagnosed a scope-magnitude bug where the production forecaster returned 10-100x over-magnitude predictions on filtered queries. Designed the replacement: scope-conditional LightGBM with leave-one-out target encoding for high-cardinality filter dimensions, stratified validation, and conformal 90% PI.
- ●Reduced CI/CD deployment time by 80% by adding affected-task builds, Docker layer caching, and parallel GitHub Actions workflows.
- ●Built the backend APIs and frontend serving all four models, loading cached S3 artifacts for sub-second inference instead of per-request retraining.
Industry Advisory Panel
ACTIVEUniversiti Teknikal Malaysia Melaka (UTeM)
Dec 2025 - Nov 2027
- ●Provide strategic guidance on curriculum development to ensure alignment with industry trends and requirements
Co-Founder & CTO
Cynco
July 2024 - Feb 2026
- ●Solo-architected production B2B SaaS platform using React/TypeScript frontend, Go/Node.js backend with Drizzle ORM, and PostgreSQL database
- ●Designed extensible payment architecture enabling new providers without major refactoring, by building a modular gateway system using factory pattern
AI Engineer
Pandai
Aug 2024 - March 2025
- ●Lead a team of five people while managing three AI products development
- ●Developed and deployed chatbot APIs and AI responses for Virtual Merdeka Hunt
Full Stack Developer
Revenue Monster
Sep 2023 - July 2024
- ●Enhanced an existing customer support chatbot by employing the Electra Small model, fine-tuned through knowledge distillation from Electra Large
- ●Researched an improved new model that produce better performance by 32% while also improving efficiency by 33%
Full Stack AI Developer
Ellzaf
Jan 2023 - Aug 2023
- ●Developed and maintained web application and AI using Django, Python, PostgreSQL, PyTorch, JavaScript, HTML, CSS, and AWS
- ●Reduced 90% of AI model inference time by replacing DistilBERT with ELECTRA-Small
Data Scientist Intern
Aibots MY
Oct 2020 - Apr 2021
- ●Researched road safety related issues for Malaysian Institute of Road Safety Research (MIROS) by utilizing the Twitter API to collect tweets from across Malaysia
- ●Orchestrated pre-processing tweets flow with noise classifier model for cleaner text data
Capabilities
PyTorchTensorFlowLightGBMCatBoostXGBoostNLPComputer VisionRAGFastTextKnowledge DistillationSynthetic DataConformal PredictionBayesian ModelingLLM AgentsLangChainDeepAgentsAgnoDeepSeekMiniMaxGLMKimiPyTorchTensorFlowLightGBMCatBoostXGBoostNLPComputer VisionRAGFastTextKnowledge DistillationSynthetic DataConformal PredictionBayesian ModelingLLM AgentsLangChainDeepAgentsAgnoDeepSeekMiniMaxGLMKimiPyTorchTensorFlowLightGBMCatBoostXGBoostNLPComputer VisionRAGFastTextKnowledge DistillationSynthetic DataConformal PredictionBayesian ModelingLLM AgentsLangChainDeepAgentsAgnoDeepSeekMiniMaxGLMKimiPyTorchTensorFlowLightGBMCatBoostXGBoostNLPComputer VisionRAGFastTextKnowledge DistillationSynthetic DataConformal PredictionBayesian ModelingLLM AgentsLangChainDeepAgentsAgnoDeepSeekMiniMaxGLMKimiPyTorchTensorFlowLightGBMCatBoostXGBoostNLPComputer VisionRAGFastTextKnowledge DistillationSynthetic DataConformal PredictionBayesian ModelingLLM AgentsLangChainDeepAgentsAgnoDeepSeekMiniMaxGLMKimiPyTorchTensorFlowLightGBMCatBoostXGBoostNLPComputer VisionRAGFastTextKnowledge DistillationSynthetic DataConformal PredictionBayesian ModelingLLM AgentsLangChainDeepAgentsAgnoDeepSeekMiniMaxGLMKimiPyTorchTensorFlowLightGBMCatBoostXGBoostNLPComputer VisionRAGFastTextKnowledge DistillationSynthetic DataConformal PredictionBayesian ModelingLLM AgentsLangChainDeepAgentsAgnoDeepSeekMiniMaxGLMKimiPyTorchTensorFlowLightGBMCatBoostXGBoostNLPComputer VisionRAGFastTextKnowledge DistillationSynthetic DataConformal PredictionBayesian ModelingLLM AgentsLangChainDeepAgentsAgnoDeepSeekMiniMaxGLMKimiPyTorchTensorFlowLightGBMCatBoostXGBoostNLPComputer VisionRAGFastTextKnowledge DistillationSynthetic DataConformal PredictionBayesian ModelingLLM AgentsLangChainDeepAgentsAgnoDeepSeekMiniMaxGLMKimi
PythonDjangoFastAPIGolangNode.jsPostgreSQLDrizzle ORMNext.jsReact Router v7Tanstack StartTanstack QueryHatchetSSEAuth.jsAlpaca APITwelve Data APITelegram BotsPythonDjangoFastAPIGolangNode.jsPostgreSQLDrizzle ORMNext.jsReact Router v7Tanstack StartTanstack QueryHatchetSSEAuth.jsAlpaca APITwelve Data APITelegram BotsPythonDjangoFastAPIGolangNode.jsPostgreSQLDrizzle ORMNext.jsReact Router v7Tanstack StartTanstack QueryHatchetSSEAuth.jsAlpaca APITwelve Data APITelegram BotsPythonDjangoFastAPIGolangNode.jsPostgreSQLDrizzle ORMNext.jsReact Router v7Tanstack StartTanstack QueryHatchetSSEAuth.jsAlpaca APITwelve Data APITelegram BotsPythonDjangoFastAPIGolangNode.jsPostgreSQLDrizzle ORMNext.jsReact Router v7Tanstack StartTanstack QueryHatchetSSEAuth.jsAlpaca APITwelve Data APITelegram BotsPythonDjangoFastAPIGolangNode.jsPostgreSQLDrizzle ORMNext.jsReact Router v7Tanstack StartTanstack QueryHatchetSSEAuth.jsAlpaca APITwelve Data APITelegram BotsPythonDjangoFastAPIGolangNode.jsPostgreSQLDrizzle ORMNext.jsReact Router v7Tanstack StartTanstack QueryHatchetSSEAuth.jsAlpaca APITwelve Data APITelegram BotsPythonDjangoFastAPIGolangNode.jsPostgreSQLDrizzle ORMNext.jsReact Router v7Tanstack StartTanstack QueryHatchetSSEAuth.jsAlpaca APITwelve Data APITelegram BotsPythonDjangoFastAPIGolangNode.jsPostgreSQLDrizzle ORMNext.jsReact Router v7Tanstack StartTanstack QueryHatchetSSEAuth.jsAlpaca APITwelve Data APITelegram Bots
AWSAWS S3DockerCI/CDGitHub ActionsGrafanaPrometheusPgBouncerPostgreSQL HAEncrypted BackupsAWSAWS S3DockerCI/CDGitHub ActionsGrafanaPrometheusPgBouncerPostgreSQL HAEncrypted BackupsAWSAWS S3DockerCI/CDGitHub ActionsGrafanaPrometheusPgBouncerPostgreSQL HAEncrypted BackupsAWSAWS S3DockerCI/CDGitHub ActionsGrafanaPrometheusPgBouncerPostgreSQL HAEncrypted BackupsAWSAWS S3DockerCI/CDGitHub ActionsGrafanaPrometheusPgBouncerPostgreSQL HAEncrypted BackupsAWSAWS S3DockerCI/CDGitHub ActionsGrafanaPrometheusPgBouncerPostgreSQL HAEncrypted BackupsAWSAWS S3DockerCI/CDGitHub ActionsGrafanaPrometheusPgBouncerPostgreSQL HAEncrypted BackupsAWSAWS S3DockerCI/CDGitHub ActionsGrafanaPrometheusPgBouncerPostgreSQL HAEncrypted BackupsAWSAWS S3DockerCI/CDGitHub ActionsGrafanaPrometheusPgBouncerPostgreSQL HAEncrypted Backups
Raspberry PiArduinoESP32MQTTRaspberry PiArduinoESP32MQTTRaspberry PiArduinoESP32MQTTRaspberry PiArduinoESP32MQTTRaspberry PiArduinoESP32MQTTRaspberry PiArduinoESP32MQTTRaspberry PiArduinoESP32MQTTRaspberry PiArduinoESP32MQTTRaspberry PiArduinoESP32MQTT
FlutterReactTypeScriptJavaScriptFlutterReactTypeScriptJavaScriptFlutterReactTypeScriptJavaScriptFlutterReactTypeScriptJavaScriptFlutterReactTypeScriptJavaScriptFlutterReactTypeScriptJavaScriptFlutterReactTypeScriptJavaScriptFlutterReactTypeScriptJavaScriptFlutterReactTypeScriptJavaScript
AI & Machine Learning
Backend & APIs
Cloud & DevOps
Hardware & IoT
Mobile & Frontend
Featured Work
[01]
AI Trade Agent Harness
Local AI trading harness that converts company research, world context, journal memory, and 5-minute OHLCV into allocator targets, risk-checked order intents, and broker reconciliation.
- ●Built bounded LLM research pipelines with provider search caps, ticker disambiguation, source-quality quarantine, prompt/version traces, retryable schedules, and durable PostgreSQL state across 53 trading and agent tables.
- ●Engineered deterministic trading safety in Go, including stale-data rejection, cash-only sizing, position caps, turnover limits, lifecycle checks, Alpaca reconciliation, encrypted backups, and broker activity journaling.
- ●Built a model-competition layer where DeepSeek, MiniMax, GLM, and Kimi run independent shadow portfolios, score their decisions against market outcomes, and identify the strongest model for operator-approved promotion into the next trading cycle.
Python ·Go ·PostgreSQL ·Hatchet ·DeepSeek ·Agno ·Alpaca ·Twelve Data ·Telegram
[02]
RizqRadar
Multi-agent research platform covering orchestration, validation, synthesis, and report delivery end-to-end.
- ●Built resilient AI Agents pipelines with multi-provider fallback logic, refusal or error handling, and validation checkpoints, improving reliability and safety of automated research outputs.
- ●Designed agent orchestration at scale with modular executors and workflow execution via Hatchet, enabling parallelizable, maintainable agent workflow.
Docker ·Tanstack Start ·LangChain ·DeepAgents ·PostgreSQL ·AWS ·Hatchet
[03]
GetDoa
Led a three-person team to develop a comprehensive Islamic prayer platform. Managed backend, database, machine learning, and deployment components.
Docker ·Django ·PyTorch ·Next.js ·PostgreSQL ·AWS ·Vercel
[04]
TriageAI
Improved an existing method by substituting CNN with Transformer Encoder, demonstrating improved F1-score by 3% for ED Triage implementation.
PyTorch ·Django ·Transformer
[05]
Protecc
A wearable care system for elderly using LSTM for Human Activity Recognition. Utilized TensorFlow Lite for Raspberry Pi optimization with real-time monitoring via Flutter app.
Python ·TensorFlow Lite ·Flutter ·Firebase ·MQTT ·Raspberry Pi ·ESP32
[06]
Elpida
Led development of two robots and mobile apps for elderly care. Implemented TensorFlow Lite and XGBoost for real-time human pose estimation and activity monitoring.
TensorFlow Lite ·XGBoost ·TypeScript ·Raspberry Pi ·Firebase
Recognition
Education
Master of Science[INCOMPLETE]
Computer Science - Intelligent System Techniques
Universiti Sains Malaysia · Research Mode
Jan 2022 - Jan 2024
Bachelor of Computer Science
Computer Science - Artificial Intelligence
Universiti Teknikal Malaysia Melaka
→ CGPA: 3.65 / 4.0
Sep 2017 - Apr 2021
Awards
Gold Award
Jejak Inovasi UTeM 2020
2020
1st Place
Artificial Intelligence Workshop 2
2020
2nd Place
Autonomous FV 2018
2018
Key Certifications
→AI in Healthcare Specialization (Stanford Online)
→AI For Medical Treatment (DeepLearning.AI)
→AI for Medical Diagnosis (DeepLearning.AI)
→Complete Flutter Development Bootcamp (Udemy)
Contributions
Open Source
Hazqeel09/ellzaf_ml[PACKAGE]
Created python package for ML implementations based on research papers
Huggingface Models[MODELS]
5 fine-tuned text classification models with 1,300+ downloads
Malaysian Dataset[DATASET]
Scraped three websites to collect over 100MB of text data for Malaysian LLM
Unifyai/ivy[CONTRIBUTION]
Wrapped PyTorch function inside Ivy for ML Frameworks unification
Publications
Revolutionizing Emergency Medicine and Triage Systems through Artificial Intelligence
Aug 2024
Data Gathering and Preparation for Social Media Data Utilisation in Road and Traffic Safety
Nov 2021