I turn slow, manual review work in banks and financial firms into AI systems that run in production. Seven years in software, from computer vision prototypes to a Philippine digital bank, and now building multi-agent LLM systems with the evaluations that keep them honest.
class Reljod:
role = "AI Engineer Β· Senior Full-Stack Engineer"
based_in = "Philippines π΅π (open to work anywhere)"
focus = ["multi-agent LLM systems", "document AI", "KYC & compliance", "GraphRAG"]
principle = "Ship, measure, iterate. Show the artifact, not the title."
background = "BS Electronics Engineering Β· Licensed ECE"
off_hours = ["calisthenics π€Έ", "property development ποΈ", "reading AI research π"]π§ͺ Reljod Labs: live at reljod.com
| Demo | What it does |
|---|---|
| π Document Compliance | Pre-screens a banking onboarding pack against a compliance rulebook. |
| π Ledger, the cat banker | Answers questions about me by taking my About page apart. |
| π¦ Teller | A Specialist agent for pension and trust operations: runs review pipelines, asks for what's missing. |
| πΈοΈ Client Manager | Builds a wealth client's knowledge graph in Postgres and answers by walking it (GraphRAG). |
All demos run real models on synthetic data. On the bench: in-house KYC automation, a name matching engine, tampering and illegibility detection, and more.
- πΈ Helped replace a third-party KYC vendor with an in-house AI solution at a digital bank: -90% KYC cost, +7 pts approval rate, +8 pts automation.
- π§Ή Designed a multi-agent eligibility pipeline that screens out 95 to 97% of ineligible records before a single LLM report is generated.
- π Shipped 20+ AI proofs of concept, from RAG and GraphRAG to OCR, computer-use and voice agents.
- πΎ First AI system ever: a computer vision line caller for tennis (OpenCV, YOLO, PyTorch).
- kita-agents.dev: AI assistants for Filipino small businesses and government services, in Taglish. Open framework: Kita-PH (FastAPI) + kita-ph-ui (Next.js).
- reljod.com: more financial-services demos, one bench item at a time.
Have manual review work that should be an AI pipeline? Let's talk at reljod.com β





