Day 4 — Next.js, Supabase, shadcn: A Day of Foundations and Clarity
Between sprint planning, agent architecture, and project setup — a day where everything finally aligns.
common.readingTime
build-in-publicagent-architectureproductivity
The Project
At Réfugiés.info, we're building a writing copilot: a tool that can take raw data (training offers, resources, etc.), filter it, rewrite it in clear language, and hand it to the editorial team for validation before publication.
The idea? AI assists, humans decide.
After an exploratory hackathon and discovering Letta (AI agents with persistent memory), we're moving into building mode. Two weeks for a POC, targeting a solid version by March 2026.
The Journal
Here, we share behind the scenes of the project: technical choices, architecture decisions, struggles, discoveries. No polish, just the reality of building in public.
Every day (or almost): an article, a perspective, a lesson.
Recent Articles
Articles
Day 3 — Letta in Action: When Stateful Memory Becomes Real
Letta demo with the editorial team: stateful memory, custom interface, and the moment the project takes shape. Réfugiés.info build in public.
common.readingTime
build-in-publiclettamemory-blocks
Day 2 — Building Data Architecture and AI Frugality with Letta
Data architecture, intelligent ingestion, and AI frugality: how Letta optimizes tokens and creates a stateful copilot for Réfugiés.info
common.readingTime
build-in-public
Day 1 — Starting Fresh to Build a Real AI Copilot
Why we're abandoning the old back-office to build an AI tool designed for Réfugiés.info's editorial team
common.readingTime
build-in-publicproductia
Day 0 — When a Hackathon Becomes a Real AI Adventure
From the frustration of 'almost there' to discovering Letta — behind the scenes of Réfugiés.info's writing copilot
common.readingTime
build-in-publicaieditorial