What I bring to an AI team
Prompts written to a spec
System prompts, examples and structured JSON output, written against a task spec and tested across GPT, Claude, Gemini and open-source models.
Evals that find the failures
Real user questions scored for accuracy, citations and tone. Every failure becomes a prompt constraint, a new test case or a guardrail.
Wired into real systems
RAG, voice and automations connected with Python, APIs, n8n and Make. Deployed and monitored, not left in a notebook.
MADA SARIMA, an AI assistant a government agency uses every day
A Malay-language RAG and voice assistant for the field officers of Lembaga Kemajuan Pertanian Muda (MADA). Officers ask by text or voice and get a cited answer from agency documents, read aloud if they want.
I built it without a dev team, directing Claude Code and Codex from architecture to deployment, and I own the prompts, retrieval and evaluation loop. It runs 24/7 on AlphaEdge (TM Credence), with weekly demos for agency leaders.
Unhelpful answers, hardest question type
Share of answers officers rated thumbs-down
The failures behind the 40% were hallucinated policy dates, retrieval vocabulary mismatch and schema drift. Each one became a prompt constraint, an eval case or a guardrail.
Ask
An officer types or speaks a question in Malay.
Faster WhisperRetrieve
Semantic search finds the right agency documents.
RAGFlow, pgvectorAnswer
A cited answer in plain Malay, shaped by the prompt spec.
OpenAI APICheck
Scored for recall, citation accuracy and Malay quality.
Eval loopSpeak
Read aloud, with a fallback voice if one fails.
Edge TTS, ElevenLabs
Screens aren't shown because it's a government system behind single sign-on. I'm happy to walk through it in an interview.
Shipped and in use
-
BrushQuote
AI estimates for painting contractors. Voice notes and site photos become a priced PDF proposal.
iOS app, US market App Store ↗ -
Al-Mathurat
Morning and evening dhikr with Arabic text and Malay translation, published and run solo.
5K+ downloads, 4.9★ App Store ↗Google Play ↗ -
Restaurant WhatsApp bot
Takes orders, answers repeat questions and broadcasts the daily menu for a restaurant in Bangi.
2 hrs saved a day -
Weekly reporting pipeline
An n8n pipeline merges four spreadsheets into the management report and emails it at 7:30 every Monday.
3 hrs saved a week
15+ years in product, engineering and leadership
- 2024 – nowAeronerve (Technerve Technology Solutions)Product & Project Manager, AI Workflow & Systems Builder
- 2025 – nowIndependent products & Haaziq.devSolo builder and automation engineer
- 2017 – 2024Fix & Go IT Specialist / Definite SecurityCEO & Digital Transformation Lead, CTO before that
- 2013 – 2017DesaKreativCo-Founder & Application Engineer
- 2010 – 2011Hewlett Packard Enterprise ServicesTechnical Solutions Consultant, Korean market
BEng, Computer & Information Engineering, Inha University, South KoreaProfessional Technologist (Ts.), MBOTCCDA, CISAM
