Personal archive / 001Software & AI systems

Open the folder to begin with sound.

Software engineering / Applied intelligence

Peru Working across systems

Diogo Abregu

The person behind the systems

Building intelligent systems
for real-world problems.

Systems Engineering student at UNI, Peru. I build software, AI agents, backend systems, and integrations for real business workflows.

Index / Selected work

Inside the archive.

Real problems. Code to explore.
Follow the thinking behind each system.

01MeowlabAgent systems / Selected work

Meowlab

AI agents for small-business workflows

Role
Software & AI development
Status
In development
Implementation outline
  1. WhatsApp events
  2. Role-based routing
  3. Conversation history

Context

Small businesses juggle quotes, invoicing, sales, and stock across disconnected tools. Meowlab brings specialized AI agents into those operational workflows through WhatsApp and connected services, with a first focus on Peruvian MYPES.

Implementation

I am developing agents for quotations and electronic invoicing through Meowlab / FacturaYa. The public runtime includes role-based message routing, voice transcription, and tenant-scoped conversation history.

Project scope

  • Specialized agents for quotations and electronic invoicing through WhatsApp.
  • Business workflows that connect sales, inventory, and external services.
  • Explicit responsibilities and coordination between agents, with attention to authorization boundaries.

System diagram / 01

Manolo · invoicing architecture

Architecture snapshot / 06 Nov 2026

From a WhatsApp message to an authorized fiscal operation.

Manolo invoicing architecture Messages flow from the user through WhatsApp, a Cloudflare Worker and a Durable Object to the Manolo orchestrator. It calls an authorized Node bridge and FacturaYa, which communicates with SUNAT. R2 stores temporary media, D1 holds conversation state, PostgreSQL supports authorization, and Luna and Decisions API provide AI inference. MessagesHTTPS webhookQueue by contactOne consolidated turnHMAC + timestamp + IDAuthorized APIElectronic invoiceInferenceRead / save statePrivate SQLTemporary media User Text · photos · documents WhatsApp Cloud API Meta · inbound & replies Cloudflare Worker Webhook · Meta signature Durable Object Groups text + images · 3 s Manolo orchestrator Intent · roles · confirmation Node bridge + Caddy Company & operation authorization FacturaYa API Drafts · issuance · PDF/CDR SUNAT Fiscal validation via FacturaYa AI · Luna / Decisions API Visual context & intent Cloudflare D1 History · checkpoint · dedup Private PostgreSQL Roles · permissions · multi-tenant Cloudflare R2 Temporary media

Scroll sideways to view the full diagram

The orchestrator runs inside the Worker. Responses and documents return through Meta Graph API; SUNAT is reached via FacturaYa. D1 remains the conversation runtime while PostgreSQL receives shadow messages.

System design & direction

Platform architecture in development: Next.js, FastAPI, WhatsApp, and Convex/PostgreSQL depending on the module, with VPS and Caddy for service operation.

Reference scenario: Sales Cat ↔ Stock Cat, coordinating sales and inventory responsibilities through business events.

The public repository also contains the Astro landing and a TypeScript WhatsApp runtime; the platform design and the repository reference represent different parts of the work.

  • AI agents
  • WhatsApp
  • Backend
  • Automation
02CapictiveCivic technology / Selected work

Capictive

Civic technology powered by data and AI

Role
Founder & AI Solutions Engineer
Period
Jan — Apr 2026
Status
Built
Implementation outline
  1. Public information
  2. Structured extraction
  3. Web experience

Context

Electoral information is extensive and scattered across public documents, candidate profiles, and interviews. Capictive makes that material easier to access and understand through software, structured data, and conversational experiences.

Implementation

I built the electoral-information platform with Next.js and Cloudflare D1, and developed ingestion pipelines using web scraping, APIs, and Gemini to extract information into structured JSON.

Candidates covered
35
Candidate CVs processed
~5,000
Users reached
~10,000

Project scope

  • Collection and processing of electoral documents and candidate information.
  • Structured extraction to turn long documents into usable application data.
  • Web and conversational experiences for exploring political and public information.

Semifinalist · Voces Credicorp 2025

System diagram / 02 · Proposed

Capictive Lima · target architecture

Architecture direction

A proposed civic-information workflow, from public sources and editorial review to district-level exploration.

Capictive Lima proposed architecture Citizens access the public portal, which requests civic data through a domain API. Editorial staff and contributors submit changes through editorial workflows. Public sources feed ingestion, which preserves evidence and submits drafts. The portal, API, civic database, editorial panel and evidence storage are proposed; ingestion is phased. reviewssubmits changesproposes draftsexploresrequests datareads & recordsdownloadspreserves originalsloads draftslinks evidence Editorial team Reviews and publishes PROPOSED Editorial panel Next.js + TypeScript Contributors Evidence-based input Citizens Browse by district PROPOSED Public portal Next.js + TypeScript PROPOSED Domain API FastAPI · Python PROPOSED Civic data PostgreSQL + PostGIS Public sources Plans & official data PHASED Ingestion & extraction Workers · Python PROPOSED Evidence storage S3 or Cloudflare R2

Scroll sideways to view the full diagram

This diagram describes a proposed direction, not a completed deployment. The observed starting point is a minimal Astro landing; manual curation comes first, with automated ingestion planned in stages. The earlier electoral product used Next.js, Cloudflare D1, Python and Gemini.

  • Next.js
  • Python
  • Gemini
  • Cloudflare D1

Engineering focus

Thinking beyond the interface.

AI & agentic systems

LLM integrations, tool calling, structured extraction, and agent workflows. Exploring retrieval and evaluation as part of building dependable systems.

OpenAI API · Gemini · Agents · RAG foundations

Backend & integrations

Python and FastAPI services, REST APIs, and webhooks that connect external systems with operational workflows and structured business data.

Python · FastAPI · REST APIs · SQL

Cloud & infrastructure

Application deployment on Linux/VPS and serverless infrastructure, with Docker, AWS cloud foundations, and attention to service operation.

AWS · Docker · Linux/VPS · Cloudflare D1

Product engineering

Turning operational requirements into full-stack applications, with maintainable architecture, clear interfaces, integrations, and iterative delivery.

Next.js · Astro · TypeScript · Git / CI/CD

Experience

Built around real workflows.

Consulting

Jul 2026 — present

M+S Metodologías + Soluciones

Junior Consultant

  • Developed an end-to-end marketplace with Python/FastAPI, Mercado Pago integration, operational automation, and VPS deployment.
  • Implemented a multi-agent development workflow with specialized roles, shared documents, and webhooks for task coordination and integrations.

Internship

Nov 2025 — Apr 2026

Growth Rockstar

Programming & Tools Intern

  • Developed reporting automation and CRM/data-management tools for internal operations.
  • Provided technical and functional support for the implementations and their use by the team.

Project work

Mar — Dec 2025

Peri Institute

Process automation

  • Built workflows for sales registration and payment validation.
  • Connected internal applications and reporting using n8n, AppSheet, and Google Apps Script.

About

Curiosity, with a practical purpose.

I work at the intersection of software engineering and applied AI: from APIs and data pipelines to agents that connect with tools and business workflows. I care about the architecture behind the interface, the boundaries of each tool, and keeping a system understandable beyond the demo.

Alongside my studies at UNI, I build products and automation for operational needs in Peru. My work spans business software, civic technology, public procurement, and conversational systems — turning complex information and processes into something people can use.

Certifications

Community

LEAD UNI

Academic Excellence Area Director

Apr 2025 — Feb 2026

Community leadership and development of the institutional website and a student job board.

Languages

Spanish
Native
English
C1 · EF SET

Education & learning

Universidad Nacional de Ingeniería

Systems Engineering

2022 — present

Undergraduate degree in progress · Top fifth of cohort · Lima, Peru

Expected graduation · December 2027

Pontificia Universidad Católica de Chile

Student exchange

Aug — Dec 2025

Software architecture, system design, design patterns, and clean code.

AWS re/Start

Cloud foundations training

2026

Cloud infrastructure foundations: EC2, S3, IAM, networking, and service operation.

Contact

Let's build something useful.

Have an engineering challenge or an idea to explore? Let's talk.