The Red Thread: The Journey That Led to This Moment
How infrastructure built for something else found its purpose in a small commune's elections
The Infrastructure That Had No Name
This story does not begin with civic tech. It begins with 3D objects on a blockchain.
In October 2023, at the Encode MultiversX Hackathon, a small team — Satish in India, Max in Ukraine, Calin, and me in Brittany — built Locki: a platform for minting and interacting with 3D Data NFTs using the Itheum protocol. We placed fourth. The code worked. The vision was ambitious. And then, as happens, life scattered the team across time zones and competing commitments.
But Locki kept growing. Smart contracts. Agent frameworks. N8N orchestration workflows for automating data pipelines. Docker deployments. A Docusaurus documentation site. ComfyUI experiments with lip synchronization. ETL pipelines. Redis coordination patterns. Provider failover chains. Each piece built because the previous piece demanded it. None of it pointed at municipal elections.
For three years, Locki was a laboratory without a destination — infrastructure assembled by curiosity, not by plan. There was no citizen chatbot on the roadmap. No notion that four thousand scanned PDFs of arretes from a Breton coastal town would one day flow through pipelines designed for 3D NFTs.
Infrastructure waits. It does not know what it is for until someone asks the right question.
The People Who Asked
The question didn't come from one person. It arrived in fragments, carried by people who each saw something different in the machinery.
Sylvain brought the idea. He saw the local elections approaching in Audierne-Esquibien and imagined what would happen if the infrastructure — the agents, the orchestration, the tracing — was pointed at civic transparency instead of abstract experimentation. The connection between the tools and the territory was his spark. Without it, the plumbing would have stayed plumbing.
Rashid brought persistence. Not the flashy kind — the kind that shows up day after day, that keeps the thread alive when momentum fades, that reminds you why you started when analysing races after races. Projects like this survive not because of brilliant decisions but because someone refuses to let them not follow the goal quietly.
Satish, who had built Locki with us from day one — on-chain 3D objects, the MultiversX smart contracts, the Next.js frontend — brought his submarine way of working. You don't always see Satish on the surface. Months would pass between messages on Discord: "Sorry, I was occupied with lot of work." Then he'd surface: "We can connect this weekend." Always there, below the waterline. When his other startup fell apart over equity conflicts — "They did not stick to what we discussed after working so hard" — that lesson shaped everything we built next. His experience with broken promises taught me to write the COLLABORATION_ADDENDUM before the first line of hackathon code. The collaboration charter exists because Satish learned, the hard way, what happens without one.
Victor brought idealism — and a courage that only career switchers know. A junior data engineer, one year of experience, coming from healthcare communications, joining his first ever hackathon. On the first day he messaged: "Excited to join the team and get started!" Then he spent weeks fighting PDF library dependencies — pdf2ocr, Tabula, pypdf, each one blocking harder than the last. Victor didn't just write crawlers. He believed in what the crawlers were for. His way of putting things — direct, hopeful, convinced that the effort mattered — carried the project through its hardest weeks. When the OCR wall finally fell, it fell because Victor had held the line long enough for us to find the pivot.
Max brought strategy. When I shared the Encode hackathon announcement in our Discord, Max didn't just say "good luck." He wrote a precise analysis of civic tech in France — the think tanks like La Fing and Institut des Futurs Souhaitables, the Tech For Good movement, how every tech disruption cycle revives the same decade-old ideas about citizen participation. "I see most of these organizations as platforms that can connect to an audience of entrepreneurs and decision-makers." Max followed the hackathon remotely, asked to try the beta, and saw what the rest of us were too deep in code to see: that the lighthouse needs to be visible, not just lit. SEO, social media, the craft of making information reach the public who doesn't know they need it yet.
And Meher — a second-year undergraduate student, jet-lagged on the US East Coast, who introduced himself with "I am really curious about your guys' backgrounds" and then said "I trust you!" and started building. He helped shape the early architecture of Forseti, brought ML intuition to the agent design, and showed that a student could contribute alongside professionals. Life pulled him away before the hackathon finished — that's life, and hackathons are brutal on schedules — but the foundation he helped lay mattered.
None of them set out to build civic AI. Each brought a piece — an idea, persistence, submarine steadiness, idealism, strategic vision, trust — and the pieces assembled into something none of us had planned.
The School Conflict That Lit the Fuse
The Pierre-Le Lec renovation project was the catalyst — not the origin, but the moment when the abstract became concrete.
Consolidating two public schools on a historic waterfront site in Audierne. Technical diagnosis completed, architect selected, funding through the Petites Villes de Demain program, students already relocated. Then the December 2025 council meeting turned heated. The majority defended the project. The opposition pointed to cost risks. Citizens watching from the gallery had no way to verify either narrative.
The information existed — scattered across deliberations, architectural reports, subsidy frameworks, press coverage. Public but functionally invisible. That evening, staring at the council minutes, the infrastructure stopped waiting. It had found its question: what if citizens could simply ask — and get a factual, sourced answer?
From Plumbing to Platform
The original sketch was modest: a knowledge base on GitHub, a chatbot, five hundred euros with open-source tools, four weeks. The Encode "Commit to Change" AI Hackathon gave it urgency. Social Impact track. Four weeks to deliver.
What followed was controlled chaos. We scraped six years of municipal publications — over four thousand arretes, council deliberations, commission documents. Victor fought the OCR wall. Meher built Forseti and the Opik integration. The N8N workflows, designed years earlier for abstract orchestration, suddenly had a concrete job: route citizen contributions from Framaforms through charter validation to public GitHub discussions.
The provider failover chain — Ollama locally for sovereignty, cloud models as backup — wasn't built for elections. It was built because we were interested in resilience patterns. But when citizens started asking questions about their town, the fact that their data stayed on a local server in France wasn't an engineering curiosity. It was a civic commitment.
We won the hackathon. Social Impact. Not for the technology — others had fancier architectures. For the commitment: that this was real. Real contributions from real citizens in a real electoral campaign.
The Platform and the 36
In December 2025, we launched audierne2026.fr. Anonymous forms through Framaforms. Public discussions on GitHub. Full traceability. Every contribution validated by Forseti against a charter of participation.
Thirty-six citizens contributed. Out of 3,600 inhabitants. One percent.
But look at what those thirty-six voices contained. A resident who knew the ANAH subsidy frameworks better than any campaign team. A parent who identified a gap in after-school transport that no candidate had addressed. Five contributions on environmental concerns — water quality, nesting birds, waste management — grounded in specific, local knowledge.
Housing dominated with eleven contributions. Schools, the economy, associations, and the environment each drew five. The programme that grew from these seeds was not written for citizens. It was written by them.
The Week Everything Converged
March 9, 2026. Six days before the first round. Four lists had published their programmes — some as structured documents, others as Facebook screenshots, Instagram stories, PDF flyers slipped under doors. Sixty files scattered across platforms.
In one afternoon, we built the bridge. Mistral OCR processed every image. ChromaDB indexed the vectors locally. Five voices entered the same index: the four electoral lists and the participatory programme from audierne2026.fr.
For the first time, a citizen could type "Que proposent les listes pour le port ?" and receive a neutral, sourced comparison of what every list actually promised. Not what a journalist summarized. Not what a campaign claimed. The actual words, from the actual documents.
The RAG adventure had begun. The windshield was clean. The gazetteer guarded the Breton names. The conversation loop let citizens ask follow-ups. And the lighthouse stood ready.
Four Lists, One Commune
Five days from now, citizens of Audierne-Esquibien choose. Four lists, four visions:
Construire l'Avenir, led by Florent Lardic — first to declare, pragmatic, rooted in sociodemographic analysis. Passons a l'Action !, led by Didier Guillon — financial rigor and municipal management. S'unir pour Audierne-Esquibien, led by Michel Van Praet — unity, sustainability, citizen listening. Cap sur Notre Futur, led by Eric Bosser — building on his experience as maire delegue of Esquibien.
Each list deserves to have its proposals read, compared, and understood on equal terms. The lighthouse does not rank. It does not recommend. It retrieves, it cites, it presents side by side. The choice belongs to the citizens.
What the Journey Taught Us
Three years of infrastructure. Not three years of civic tech — three years of building tools that didn't yet know their purpose. From 3D NFTs on MultiversX to municipal PDFs in Audierne. The civic application emerged not from a master plan but from a convergence: the right people, the right moment, the right conflict, the right stubborn refusal to let the project stay abstract.
Sylvain's idea. Rashid's persistence. Satish's submarine steadiness and the equity lesson that shaped our collaboration charter. Victor's idealism on his first hackathon. Max's strategic eye connecting us to the broader civic tech landscape. Meher's early trust. The school conflict that demanded an answer. The hackathon that imposed a deadline. The infrastructure that was ready because it had been built for building's sake — even when "building's sake" meant 3D objects on a blockchain.
That is perhaps the deepest lesson. You cannot plan serendipity. But you can build plumbing that is ready when the water arrives.
Three years taught us that civic AI is not about intelligence. It is about fidelity — to sources, to neutrality, to the citizens who trust you with their questions. The hardest problem was not retrieval or synthesis. It was the discipline of saying "I don't know" when the documents don't contain an answer.
Three years taught us that one percent participation is not failure. Those thirty-six voices carried knowledge that no campaign team possessed alone.
Three years taught us that a lighthouse does not need to be large. It needs to be honest. It needs to stand on rock. And it needs to turn its beam for everyone.
The Decision
On March 15, the citizens of Audierne-Esquibien will decide. Not the AI. Not the algorithms. Not the agents. The citizens — in the privacy of a voting booth, with whatever mix of conviction, hope, frustration, and local knowledge they carry.
Our role was never to influence that decision. It was to ensure that anyone who wanted to understand had the tools to do so. That the information landscape was navigable. That the lighthouse was lit.
The beam keeps turning. The decision is yours.
Bonne chance, Audierne-Esquibien.
Related: The Lighthouse Manifesto | O Capistaine! My Capistaine! | 36 Voices Out of 3,600 | The RAG Adventure Begins