I build and operate AI systems that ship: retrieval pipelines, evaluation + quality gates, and production infrastructure for LLM-backed products.
| Area | Typical work |
|---|---|
| Retrieval infrastructure | Hybrid retrieval (graph + vector + structured filters), indexing, ranking, tuning |
| Production readiness | Load tests, latency/concurrency modeling, rollout checks, operational guardrails |
| Security automation | Repeatable OWASP ZAP scans and baseline security workflows |
| Reliability | Routing, TLS hardening, incident-ready logging/metrics, active-active patterns |
| Category | Stack |
|---|---|
| Languages | Python · TypeScript/JavaScript · Bash |
| AI/Retrieval | LangChain · embeddings pipelines · hybrid ranking · evaluation workflows |
| Data/Stores | Neo4j · PostgreSQL/pgVector · TiDB · MySQL |
| Infra/Delivery | Docker · Nginx · Linux |
| Testing/Quality | k6 · Locust · Playwright · OWASP ZAP |
Most production code I work with is confidential (state government use cases). These deployed products are publicly accessible:
| Product | Link |
|---|---|
| Dayang chatbot (Sarawak services portal) | https://service.sarawak.gov.my/web/ |
| Court-related project (public article reference) | https://ekss-portal.kehakiman.gov.my/portals/web/home/article_view/0/5/1 |
| Malaysia public library chatbot (button-based) | https://www.u-library.gov.my/portal/web/guest |
These repos mirror the way I build systems end-to-end (pipeline → retrieval → validation), even when production code isn’t public.
| Area | Repository | What it shows |
|---|---|---|
| Local multi-agent AI app | agentic-video-analyst | offline inference · multi-agent orchestration · desktop app engineering |
| Graph ingestion + retrieval | neo4j-document-pipeline | graph modeling · retrieval API patterns for LLM workflows |
| Vector + hybrid experiments | tidb-vector-llm-testbed | relevance/scoring experiments · indexing tradeoffs |
| Embedding pipeline | mysql-to-pgvector-embeddings | extraction → embeddings → pgVector semantic layer |
| Structured retrieval | faq-retrieval-system | structured query layer for grounded answers |
| Performance testing | playwright-dayang, k6-for-custom-dify | UX + API load testing approaches for assistants |
| Security automation | zap-security-api | ZAP baseline/quick/full scan exposed via API |
| Experiments | playwright-study, besu-ibft2.0 | targeted learning repos (testing + distributed systems) |
- Measured improvements (eval + monitoring) over demo-only features
- Quality, latency, and security as release criteria
- Operable systems: clear failure modes, logs/metrics, and runbook-friendly workflows