AI agents · RAG · LLM evals · MCP

Agentic AI Developer in the Philippines

Les John Paul Oliver · Urdaneta City, Pangasinan, Philippines · Remote worldwide

Agents are LLMs that do more than answer: they look things up, call your systems, and take actions. I build them for production, which means the agent only gets the tools it needs, every write is logged, and a person approves anything that touches money or records.

In production today: an approval queue where deterministic rules decide, Claude explains each flagged case, and a human confirms every action across two legacy systems. On this site, Pao answers from a scoped knowledge base and books calls through a calendar tool.

What I build

  • AI agents with tool use: lookups, record creation, drafting, and booking, limited to the actions you allow.
  • RAG: answers grounded in your own documents and data, so the agent cites what it knows instead of guessing.
  • LLM evals: test sets for real questions, off-topic and adversarial prompts, run before every release so quality doesn't silently drift.
  • MCP servers and Claude Code skills that give agents safe, scoped access to your systems.

Guardrails I ship by default

  • Scope locks so the agent declines off-topic and prompt-injection attempts.
  • Human-in-the-loop approval where money or records change, with an audit trail and a way to reverse.
  • Rate limits and a daily budget so a busy day never becomes a surprise bill.

Proof: work I shipped

AI-Assisted Refund Automation

In production

Automates a utility's daily refund and payment-correction approval queue across two legacy systems. Deterministic recipes decide, Claude explains the flagged cases, and a human confirms every action.

Read the case study

ACIRS: 24/7 AI Customer Support

In build

An AI customer-support system that answers members around the clock from the company knowledge base, hands off to the right department, and builds a customer record from every conversation.

Read the case study

Sneaker Symphony: AI-Run Marketplace

Live

A Philippine sneaker marketplace at sneakersymphony.com, built and run by one person. Yupoo and WGStores supplier scrapers, Claude Haiku product-title parser, GCash, bank transfer, and COD payments with proof-of-payment upload and per-pair QC approval.

Read the case study

Questions people ask

What is agentic AI?

AI that plans and acts through tools, not just a chatbot that replies. A useful agent is scoped to a few safe actions, logs what it does, and asks a person before anything irreversible.

Do you build RAG systems?

Yes. RAG grounds an assistant's answers in your own documents and data, so it can answer specifics about your business and say when something isn't covered.

How do you test an AI agent before launch?

With evals: a set of real questions plus off-topic and adversarial prompts, checked before every release, alongside normal automated tests for the code around the model.

How much does an AI agent project cost?

Pricing depends on scope. Book a free 30-minute call and you get a straight answer on cost and timeline.

Sa Filipino

Ano ang agentic AI?

AI na hindi lang sumasagot kundi gumagawa rin ng aksyon gamit ang mga tool na pinapayagan mo, at humihingi muna ng approval bago ang anumang hindi na maibabalik.

Pricing depends on scope

Book a free 30-minute call. You get a straight answer on cost and timeline.

Book a free 30-minute call