Engineering the
Future,
Grounded in FaithFaiq HilmanAI Systems Architect

The dots only connect looking backwards. Move forward anyway.

AI systems architect delivering production AI, ERP automation and LLM systems. Led applications on a RM 10M+ platform serving 30,000+ users and built an AI agent that migrated 880 projects into Odoo in 3 days.

Explore selected work

Building AI That
Actually Ships

I'm an AI systems architect and engineer delivering production AI, ERP automation and LLM systems across consulting and enterprise environments. I've led applications on a RM 10M+ platform serving 30,000+ users, shipped a RM 2M+ talent-evaluation system six weeks early, and built an AI agent that migrated 880 projects into Odoo production in three days.

I work full-stack across Python/FastAPI, React/TypeScript, .NET/C# and PostgreSQL, deploy on AWS, Azure and my own VPS, and specialise in RAG, multi-agent orchestration, and LLM evaluation: translating complex AI capabilities into deployed business systems.

When I'm not building, I'm teaching. An early adopter of agentic coding, I've trained 10+ engineers on Claude Code, Codex and Cursor, and designed AI training for 90+ AI champions across 8 countries.

Where I've Built

  1. Aug 2026 to PresentUnderway

    CitaGlobal

    System Architect

    Driving AI systems architecture and delivery at CitaGlobal Telco within a 20+ subsidiary group, working with the CEO, General Manager, Finance, IT, Procurement and HR.

    • Built a natural-language AI agent for Odoo ERP and used it to migrate an eight-figure portfolio of 880 telecom projects into production in 3 days, saving five figures in vendor costs
    • Cleaned the Excel source of truth, released to production in logged stages with rollback, and rebuilt the management dashboards in Odoo
    • Building the group-wide HR platform with the HR team, approved for rollout to 500+ staff, covering leave, claims, appraisals and AI-assisted recruitment that ranks CVs with evidence-linked scoring
    • Authored the group AI policy, covering model access, pricing, SLAs and acceptable use, approved group-wide, and set up the AI use-case register
    Projects migrated
    880in 3 days
    Portfolio value
    8 figures
    Vendor costs saved
    5 figures
    HR platform
    500+staff, group-wide

  2. Mar 2026 to Jun 2026

    NV5

    AI Solutions Engineer

    Built NV5 APAC's AI operating model (knowledge base, opportunity framework, backlog and intake), turning business requests and field bottlenecks from engineering and commissioning teams into scoped automation projects with named owners and ROI logic.

    • Proved in 2 hours a Power Automate and Microsoft Graph workflow scoped at 8 weeks, converting live SharePoint comments and photos into editable inspection reports to replace a 3–4 hour engineer task per inspection
    • Lifted classification accuracy on historical building-management point tags from 69.6% to 84.3% by combining OCR and ML, extracting 1,640 candidate records at 85–94% OCR accuracy under engineer review
    • Designed hands-on AI training, prompt frameworks and communications for 90+ nominated AI champions across 8 APAC countries, focused on measurable workflow automation
    Proven in
    2 hrsscoped 8 wks
    Classification accuracy
    69.6% to 84.3%
    Candidate records
    1,640
    AI champions
    90+8 countries

  3. Mar 2026 to PresentUnderway

    Radio Televisyen Malaysia (RTM)

    Independent AI Consultant, Generative Media Production

    Produced broadcast-quality generative AI sequences that aired in Episode 3 of RTM's nationally televised DIA programme. Now storyboarding the NeuroIman pilot episodes for producer review.

    • Worked directly with RTM producers, building repeatable Seedance, Kling and ChatGPT workflows that generate each sequence from a brief in a few hours
    Programmes
    2DIA · NeuroIman
    Aired nationally
    Ep 3DIA

  4. Aug 2024 to Mar 2026

    EY

    Technology Consultant, AI Engineering

    Led 3 production AI applications on a RM 10M+ enterprise platform serving 30,000+ users at a top-5 Bursa Malaysia-listed company, personally building 70–90% of each and ranking top-3 contributor in a 100K+ LOC codebase across React/TypeScript, .NET/C#, Python/FastAPI and PostgreSQL.

    • Shipped a RM 2M+ talent-evaluation system to production 6 weeks early, going from 60% complete to deployed in 3 days, with LLM and ML scoring across 2,000+ candidates and 5 competency frameworks
    • Unlocked a RM 400K milestone payment by rescuing a live executive demo, redesigning the data schema and evaluating 50+ candidates against 2 job descriptions in 2 hours; the client credited this with saving the project
    • Cut new-agent deployment from weeks to days with a reusable multi-agent framework (factory and strategy patterns) powering a RAG chatbot over 40+ knowledge sources with OCR, image, compliance and comparison agents
    • Built an AI assistant for Malaysia's electricity-generation sector, unifying 19 fragmented data sources behind natural-language queries
    • Owned CI/CD (AWS CodePipeline) and releases across 3 production systems with zero missed deadlines
    • Drove agentic coding adoption, training 10+ engineers on Claude Code, Codex, Cursor and AI-assisted workflows
    Platform
    RM 10M+
    Users
    30,000+
    Milestone unlocked
    RM 400K
    Evaluation system
    RM 2M+
    Shipped early
    6 wks

Earlier: PricewaterhouseCoopers (PwC), Assurance Associate, Mar 2020 to Jul 2020, where trend analytics on financial statements led to the recovery of US$25,000+ in variances.

What I've Built

XIIIIIIIIIVVVIVIIVIIIIXXXI Agent GatewayHealth CanvasIlmuAILLM Eval

Malaysia Agent Gateway

4 agent interfaces

Built a Malaysia-focused agent platform exposing invoice, payment, approval and compliance workflows through CLI, HTTP API, MCP and reusable agent skills. Implemented stateful Python orchestration, approval gates, event-driven payment ingestion and adapters for LHDN MyInvois, CIDB and halal-compliance workflows.

  • MCP
  • CLI
  • LHDN MyInvois
  • CIDB

Health Canvas

Top 5 among 150+ participants

Placed Top 5 as the only solo developer among 150+ participants at a hackathon backed by Malaysia's Ministry of Health, National AI Office and AI Tinkerers KL. Built an AI clinical canvas in 48 hours using React, TypeScript, FastAPI, FAISS and GPT-4, processing 15-page medical histories with 90%+ extraction accuracy through RAG.

  • RAG
  • FAISS
  • GPT-4
  • React

IlmuAI: Multilingual Islamic Knowledge RAG System

250,000+ knowledge chunks

Built and deployed a cited bilingual RAG system over 40,000+ Islamic sources, embedding 250,000+ chunks from the Quran, hadith and fiqh. Implemented pgvector and BM25 hybrid retrieval, cross-encoder reranking, LLM-as-judge validation, a streaming React UI and Docker/Caddy deployment on Hetzner.

  • RAG
  • pgvector
  • BM25
  • Cross-encoder
  • Docker

LLM Evaluation Framework

20% to 75% TinyLlama accuracy lift

Built an evaluation framework for base, API and QLoRA models with LLM-as-judge scoring, reproducible A/B logs and 87% accuracy on domain benchmarks. Fine-tuned TinyLlama-1.1B on Malay instruction data on a single 12GB GPU, raising task accuracy from 20% to 75%.

  • LLM-as-Judge
  • QLoRA
  • Fine-tuning
  • TinyLlama

How I Build

AI & LLM

  • Multi-agent orchestration
  • RAG
  • pgvector
  • BM25
  • Reranking
  • LLM-as-judge
  • Guardrails
  • Prompt engineering
  • MCP
  • QLoRA fine-tuning
  • Local LLMs

Machine Learning & Data

  • PyTorch
  • Hugging Face Transformers
  • PEFT
  • TRL
  • bitsandbytes
  • scikit-learn
  • XGBoost
  • pandas
  • NumPy
  • Embeddings
  • ETL
  • Statistics

Software Engineering

  • Python
  • TypeScript
  • JavaScript
  • React
  • React Native
  • HTML / CSS
  • Node.js
  • FastAPI
  • .NET / C#
  • SQL
  • REST APIs
  • SSE streaming

Cloud & Infrastructure

  • AWS Bedrock
  • AWS CodePipeline
  • AWS EC2
  • AWS S3
  • AWS RDS
  • Azure
  • Docker
  • CI/CD
  • Hetzner VPS
  • Caddy
  • PostgreSQL
  • Redis
  • Celery
  • MongoDB
  • SQLite

Enterprise Systems

  • Odoo 19 (JSON-2 API)
  • ERP integration
  • Microsoft 365
  • SharePoint
  • Power Automate
  • Microsoft Graph

Agentic Development

  • Claude Code
  • Codex
  • Cursor
  • Agentic coding workflows
  • Test-driven refactoring
  • Team training

Where I Learned

MSc Data Science

City, University of London

Sep 2022 to Jul 2024

Relevant Coursework

  • Machine Learning
  • Statistics
  • Big Data
  • Neural Computing
  • Knowledge Graphs

Bachelor of Commerce

University of Melbourne

Jan 2017 to Dec 2019

Relevant Coursework

  • Corporate Finance
  • Investments
  • Derivatives
  • Accounting

Credentials & Certifications

  • ISC2 Certified in Cybersecurity

  • Microsoft Security, Compliance and Identity Fundamentals

  • Microsoft Azure Fundamentals

Let's Build
Something Great

I'm always interested in hearing about new opportunities, challenging projects, or just connecting over shared interests in AI.

Get in Touch LinkedIn GitHub

Designed & Built by Faiq Hilman · 2026