WORK // 03

Systems we built. Running in production.

Build logs beat case studies. These are the three we can show today. More cases are live under NDA - ask on a scoping call.

INTERNAL // 01

AISO AI Engine

Phased over several sprints during 2025-26. Production as of April 2026.

We built the AI platform that runs AISO Hub.

A 10-stage analytics and deployment pipeline that runs daily against every AISO Hub client - export → analyze → ideate → approve → spec → implement → deploy → verify, with human checkpoints at three stages.

SCOPE
  • • API service (Fastify) exposing board tasks, clients, analytics.
  • • Pipeline of 10 skill stages (export → daily-analysis → idea-generator → idea-approver → spec-writer → spec-approver → implementer → deploy-approver → deployer → deploy-checker).
  • • Auto-approval paths for safe changes; human review for sensitive content.
  • • Service-token auth + role-based access.
STACK

TypeScript, Fastify, Postgres, Cloudflare.

WHAT IT PROVES

We run a real pipeline against real client data, daily. Not a demo.

Read the build log
INTERNAL // 02

AISO Orchestrator

6 weeks to v1, then iterative.

A multi-agent control plane with live event streams.

A Next.js + MongoDB replica-set application that watches for agent work via Change Streams, pushes events over Server-Sent Events to a Claude Code orchestrator skill, and runs a dashboard for configuration, activity, and health.

SCOPE
  • • Next.js 15 app on port 7777.
  • • MongoDB replica set (required for Change Streams).
  • • 17 configurable agent types.
  • • SSE event stream to the orchestrator skill.
  • • Entity management (clients / brands / projects) with cascading selects.
  • • Activity log, health endpoint, stale-record cleanup.
  • • Inter-agent messaging MCP - threads with priority, assignment, and cascading context.
STACK

Next.js, MongoDB, SSE, TypeScript, Claude Code skills, MCP.

WHAT IT PROVES

We build production-grade multi-agent systems - not just wrappers around a single LLM call.

Read the MongoDB Change Streams build log
CLIENT // 01

2-week production grocery scraper

14 working days.

We shipped a production grocery scraper in two weeks.

A Portuguese grocery chain needed competitive catalog data and price-change detection. We scoped, built, and shipped in 14 working days - scraping, normalization, change detection, and a simple query interface. Details under NDA.

SCOPE
  • • Scraper covering catalog pages across multiple categories.
  • • Normalization layer (SKU dedup, unit normalization, category mapping).
  • • Daily change detection (new items, price changes, out-of-stock).
  • • Query interface for the buying team.
STACK

Python, Postgres, Cloudflare, custom query UI.

WHAT IT PROVES

Narrow scope + disciplined method ships in weeks, not quarters.

Scope a project
ADDITIONAL // REFERENCES

More engagements (short).

  • • Lisbon architecture studio, 6-month retainer. Full site rebuild on Astro + Sanity, multilingual, ongoing content and feature work.
  • • Agent Messaging MCP. Open protocol implementation - threads, priorities, cascading entity context. Runs between every AISO agent.
Scope a project Free AI Readiness Audit