Blockchain & Cie
Talk about your system
Talk about your system

Build something new

You get a working product or AI agent, tested, deployed and documented, ready for your team to take over.

Do any of these situations sound familiar?

  • A team manually repeats a time-consuming business workflow.
  • An existing chatbot gives poor answers.
  • A validated idea needs to become a demonstrable product fast.
  • A dApp project needs smart contracts and a Web3 frontend.
  • An investor demo is coming up and the product needs to work.

The offers on this path

AI Agent Deployment

deploying an AI agent in production, with bounded autonomy

An AI agent in production on your business workflow, with bounded autonomy and human control. Supervised, tested, documented. Your team stops repeating the workflow by hand.

Duration
2 to 4 weeks

Triggers

  • A team manually repeats a time-consuming workflow.
  • An existing chatbot gives unsatisfying answers.
  • You need to show investors a concrete automation.

Deliverables

  • Protocol choice (WhatsApp, Matrix, API, MCP).
  • Multi-provider AI model routing with automatic failover.
  • Hardened Docker stack (isolation, secrets, restricted ports).
  • Human control rules and guardrails.
  • Monitoring, health checks and logs.
  • Unit and integration tests.
  • Operations and handoff documentation.
  • Deployment on your cloud or self-hosted.

What you provide

  • A clear definition of the business workflow to automate.
  • Access to the API or account for the chosen channel (WhatsApp, etc.).
  • Keys and budget for the AI model providers.
  • Deployment target (cloud or existing infrastructure).

Eligibility

  • A real, repeatable business workflow already exists.
  • The organization accepts human control over sensitive decisions.

Out of scope

  • Training or hosting custom models.
  • A full rebuild of an existing application unrelated to the agent.

POC / MVP / dApp Build

full build, from design to deployment

A working, deployed product, ready for an investor demo or a real user test.

Duration
POC: 4 to 6 weeks. MVP: 3 to 4 months.

Triggers

  • A validated idea (first signups or client interviews) to turn into a product.
  • An investor demo deadline.
  • A need to ship before competitors do.

Deliverables

  • Architecture and technical stack choices.
  • Frontend (React or Next.js, TypeScript, mobile-first).
  • Backend (Node.js or Python/FastAPI).
  • Smart contracts if the project is a dApp (Solidity, Hardhat, Wagmi/Viem).
  • Database and API design.
  • Unit and integration tests.
  • CI/CD and Docker deployment.
  • CLAUDE.md / AGENTS.md documentation for agentic traceability.
  • Documentation and handoff.

What you provide

  • Product brief.
  • Mockups or reference sites.
  • Necessary access keys.
  • Deployment target.

Eligibility

  • The idea is validated (early users, interviews or equivalent).
  • A decision-maker is available during the engagement.

Out of scope

  • Long-term maintenance (see AI Agent Retainer).
  • Marketing, SEO or app store submission.

Method

Building follows the same scoping approach as fixing, adapted to a new system.

  1. Scoping

    A call to define the workflow, the product and the expected outcome.

  2. Diagnosis

    Architecture and stack choices, driven by the actual need, not a default standard.

  3. Work within a locked scope

    The system is built within a fixed scope, tests and deployment included.

  4. Handoff

    Technical documentation and handover, so your team can take the lead.

Responsibilities

  • Antoine Delamare leads the engagement technically from start to finish.
  • Any AI agent deployed runs with bounded autonomy and human control over sensitive decisions.
  • The scope locked after scoping is binding for both parties.

A concrete proof

Cotton traceability, ESA feasibility study through Parametry

Feasibility study funded by the European Space Agency (ESA), run through Parametry: tracing organic cotton from field to fiber.

Problem

The project needed a full system linking satellite observation, certification and field reporting, with no existing technical reference.

Intervention

Designed and built end to end. A mapping dashboard with plot capture. A smart contract suite for certification (identity token, transaction certificates, dual-verification oracle). A dual-database backend for geospatial layers. A Python pipeline for satellite data processing. A trilingual WhatsApp agent for producer registration.

Results

~270 commits across 9 GitLab repos
Contribution volume across every repository of the project, between October 2025 and August 2026.
105 tests on the smart contract suite
Hardhat test coverage on the identity token, the transaction certificates and the dual-verification oracles.
~196 tests on the satellite pipeline
Python test coverage on the satellite data processing chain and the compliance score computation.

Limits

This proof documents a feasibility study and an end-to-end proof of concept. It does not document a large-scale production deployment or an obtained regulatory certification.

Tech stack

  • React
  • Vite
  • Solidity
  • Hardhat
  • FastAPI
  • PostGIS
  • Python

Who this is for, and who it is not

Who this is for

  • A real business need or workflow, already identified.
  • A sponsor able to approve the scope and the deployment.
  • The organization accepts human control over an AI agent's sensitive decisions.

Who this is not for

  • Training or hosting custom AI models.
  • A speculative crypto project with no demonstrated software need.

After deployment

Once the system is in production, a monthly Retainer takes over: maintenance, new bounded workflows and security, without a dedicated hire.

AI Agent Retainer

Your AI agent stays stable and up to date, and gains new workflows every month, without a full-time hire.

Included deliverables

  • Fixes and regression resolution.
  • 1 to 3 new bounded workflows per month.
  • Security review and dependency updates.
  • Monthly performance report and recommendations.
  • Asynchronous support within 24 business hours.

Frequently asked questions about this path

How long does a POC or an MVP take?

A POC takes 4 to 6 weeks, an MVP takes 3 to 4 months. The exact duration depends on the scope defined at kickoff.

Can the AI agent act alone on sensitive decisions?

No. Every deployed agent runs with bounded autonomy and human control rules, defined with you.

Do you also build smart contracts?

Yes, when the project is a dApp. Solidity, Hardhat and common Web3 libraries are in scope.

What happens if the scope changes during the engagement?

The scope is locked after kickoff. Any change is discussed and approved before it starts.

A system to build?

Describe your need and your deadline on a call. You leave with a clear recommendation, no commitment.

Talk about your systemWrite an email