Janriz Libres
Full-Stack / AI Engineer
I build production systems: full-stack platforms, AI agents, and the retrieval behind them.
About
I'm Janriz, a full-stack and AI engineer in Cebu, Philippines.
Summa cum laude in Computer Science, Xavier University.
Game dev on the side.
Janriz Libres Cebu, Philippines
Around home
At LogoLife, I built Athena, an assistant that remembers each student.
I like owning a system end to end, from the schema to the screen.
I work in TypeScript, Next.js and NestJS. For AI, the Vercel AI SDK, LangChain and MCP.
Open to full-stack and AI roles, remote.
Every star is one of my notes. The lit ones are where these lines come from.
Work
First a jump back to August 2023, where it started. Then forward through four roles to now.
Between them, the AI breakthroughs that changed what I could build.
August 2023
Home, before any of it.
The Crusader Yearbook
Software Developer · Aug 2023 – May 2024
Kept the yearbook’s Laravel site running, and wrote the OCR and mailing tools behind its photo sessions.
- Laravel
- WordPress
- Python
Nov 2023
GPT-4 Turbo
A cheaper GPT-4, and an API for building assistants.
Mar 2024
Claude 3
Anthropic’s first GPT-4-class family.
May 2024 · built with
GPT-4o
GPT-4-class at a production price. This site’s chat runs on it.
Elinnov Technologies
Software Engineer Intern · Aug – Nov 2024
Built API endpoints for a warehouse system, and client pages that lifted traffic 10% in their first month.
- ASP.NET
- React
Sept 2024
OpenAI o1
The first reasoning models.
Nov 2024 · built with
Model Context Protocol
One protocol for plugging tools into any assistant. Athena runs on it.
Feb 2025
Claude 3.7 Sonnet and Claude Code
The coding agent moved into the terminal.
Apr 2025 · built with
GPT-4.1
A million tokens of context, and better at following instructions.
Tolstoy
Full-Stack / AI Engineer · Aug – Oct 2025
Built the retrieval behind an AI shop assistant. One client’s sitewide conversion rose 11% within six days.
- TanStack
- AWS
- Vercel AI SDK
- GPT-5
- GPT-4.1
Aug 2025 · built with
GPT-5
A new flagship generation. I shipped with it at Tolstoy and LogoLife.
LogoLife
Full-Stack / AI Engineer · Oct 2025 – Oct 2026
Shipped Athena, an assistant that remembers students across sessions, and the AI Web Crawler behind it.
- TypeScript
- MCP
- Mem0
- GPT-5 mini
- Grok 4 Fast
Nov 2025
Claude Opus 4.5
State of the art at real-world software engineering.
Feb 2026
Claude Opus 4.6
Agent teams on one codebase, with a million tokens of context.
Apr 2026
Project Glasswing
A gated model found zero-days in every major OS and browser.
June 2026
Claude Fable 5
The first Mythos-class model anyone could use. It felt like AGI.
Sept 2026
Claude Opus 5.5 and Sonnet 5.5
The newest frontier models, six days apart.
Projects
What I built along the way.
AthenaAn AI assistant with a memory
An admissions assistant that coaches students and remembers each one across sessions.
React · Express · Vercel AI SDK · Mem0 · MCP
⊘ NDA. Patterns only, no screenshots.
- Counselor Dashboard Every session’s context, with Athena built in
- E2E Testing Suite Eight personas, every workflow
- Memory that lasts across sessions
- A secure MCP server for live student data
- Nudges that bring students back
AI Web CrawlerA five-stage pipeline that finds, cleans and enriches records
It searches the web, turns pages into records, and enriches them, catching duplicates twice so nothing is paid for twice.
⊘ NDA. Patterns only, no screenshots.
- Crawler Dashboard Every crawl’s health, live
- Data Maintenance Pipeline Records that keep themselves current
Start · Load the queries
MongoDB Atlas
The crawler starts by loading its search queries from the database.
Search · Search and crawl
Firecrawl API · search and web crawling
Each query goes out to the web, and every page that comes back is crawled.
Stage 1 · Extract
OpenAI GPT-5 mini · structured output
Each page becomes a record that fits one schema.
Stage 2 · Dedup, layer 1
OpenAI text-embedding-3-large · similarity check
Near-duplicates merge by embedding similarity, before anything costs more.
Stage 3 · Enrich
Grok 4 Fast · chat completion
Each record is filled in with what its page left out.
Stage 4 · Dedup, layer 2
OpenAI text-embedding-3-large · similarity check
A second check, for records that only match once they’re enriched.
Stage 5 · Save
MongoDB Atlas · document storage
The clean records are stored.
Paid for once
Two dedup layers, one before enrichment and one after it, cut enrichment costs by 70%.
⊘ NDA. Patterns only, no screenshots.
- −70% enrichment costs
On my own time
Three things I built for myself, wound into one ribbon.
Contact
Next, I want to build AI products people can trust with real work, on a team that ships.
Open to full-stack and AI roles.
It’s in Cebu.