ALEF/ONE
Centervert · Boost Capital · Philippines

Local-first agent tooling · production AI · full-stack systems

I build the shortest path from idea to shipped AI system.

I’m Alfredo Dela Peña III, engineering lead at Centervert and AI engineer at Boost Capital. I’m the guy you call when the job crosses frontend, backend, agents, evals, infra, documents, fraud checks, dashboards, memory, and whatever broke at 2AM. True harness engineer: I wire the tools, bend the workflow, and ship the system.

Email GitHub 
┌─ ALEF/ONE ───────────────────────┐
│                                  │
│  A L E F / O N E                 │
│                                  │
│          _.-''''-._              │
│       .-'    O    '-.            │
│      ╱    .-'''-.    ╲           │
│     ╱    ╱   O   ╲    ╲          │
│    │    │   /#\   │    │         │
│     ╲    ╲  ╲#/  ╱    ╱          │
│      ╲    '-...-'    ╱           │
│       '-.         .-'            │
│          '-------'               │
│              ||                  │
│             /__\                 │
│                                  │
└──────────────────────────────────┘
Work surfaces
01 / document forensics

Production document parsing, structured extraction, and fraud detection.

At Boost I work on AI Parser Studio: OCR, multi-provider LLM extraction, field dictionaries, fraud scoring, monitoring, and failover around documents that feed onboarding decisions.

02 / local memory

Memory should not require a ceremony.

Memorantado started from a simple irritation: I wanted useful local memory without logging into something just to put rows in SQLite.

03 / agent bridge

Agents need pipes, not personality decks.

Tulay ng Mama Mo is a two-way communication bridge between Codex and Claude Code. The name is what happens when I name things too quickly.

Personal inventory
01

Memorantado

I made my own local-first memory for me because I could not be bothered logging in just for some SQLite.

 02

Tulay ng Mama Mo

I am notoriously bad at naming things, so I just thought of this. It is used for two-way communication between Codex and Claude Code.

 03

Herdr

Forked terminal agent multiplexer. I care about running multiple agents without turning the terminal into a circus.

 04

Pi-Plankton

Forked Plankton because Alex Fazio could not be bothered merging my PR. Ported it for Pi and added Elixir support.

 05

Dev Booger

Logger library. The name happened because I was picking my nose for boogers while trying to think of one.
Companies and client work
Current

Boost Capital

AI engineer on AI Parser Studio: a production FastAPI document-processing platform for OCR, structured extraction, fraud detection, batch parsing, monitoring, and multi-provider LLM failover.

 Client via Boost

GCash

Client work through Boost. Worked on document parsing and fraud-detection gaps, including payslip AI-edit/inpainting cases flagged by risk teams.

 AI / full-stack

Lawlab

Worked on AI evaluation and contract/property document analysis, plus product surfaces around the workflow.

 Real estate product

Nona

Property product work around listings, recommendation surfaces, and real-estate UX.

 Community product

Hey Roomie

Real-time chat, community tooling, influencer workflows, event tracking, and responsive management interfaces.
How I tend to work
01Read

Find the actual constraint before touching code.

02Cut

Remove steps, services, and abstractions that do not carry weight.

03Check

Use evals, logs, and boring verification instead of vibes.

04Ship

Prefer the smaller system that survives contact with users.

For practical AI systems

If the problem has documents, agents, messy workflows, or production risk, email me.

alfredo.sdpiii@gmail.com github.com/alfredosdpiii Carmona, Cavite · Philippines