Mindflow
Mindflow is an ultra-automation platform for cybersecurity-oriented companies. It’s a workflow engine that orchestrates thousands of APIs and enables any type of action, including AI, at scale. I joined the project to scale the platform as an AWS cloud expert and DevSecOps. I handle everything related to infrastructure and product architecture. From 2025, the product remains an API orchestrator strongly oriented towards AI (chat/agents and others).
Technologies
Here are the main technologies and services used:
| Category | Tech/Services |
|---|---|
| Front & UI |   |
| Languages & Runtimes |   |
| Tooling & Tests |   |
| Cloud & Infra |  (Step Functions / EventBridge / Quicksight / AppSync / Bedrock / Batch / Organisations / multi-tenant) |
| Infra as Code & CI/CD |    |
| Other | Github, OpenAPIs, tenant automation, analytics parquet exports |
Challenges
-
Scalability
Going from 5 to ~30 customers and supporting load increases per client (e.g. 100 → 5000 executions/day) without degrading service quality. -
Deployment automation
Automate client deployments as much as possible (multi-tenant, sandbox, internal accounts) with safe patterns. -
Costs & optimizations
Limit unexpected costs related to massive API orchestration and intensive cloud service usage. -
Monitoring & observability
Improve monitoring to quickly detect regressions and facilitate creation of sandbox/internal/client accounts. -
Infra/product security
Many infra/sec features to implement to guarantee isolation and security between clients.
Successes
- Analytics via Quicksight and parquet file generation for ingestion and reporting.
- Refactoring of data schemas and optimization of access patterns for large audit tables.
- Implementation of a fully serverless blue/green deployment.
- Deployment of the solution on a client AWS account and creation of multi-tenant trial accounts for customer success and prospects.
- Migration to event-driven / microservices systems to replace old monolithic systems.
- Rewriting and implementation of a serverless AI Agents solution replacing dedicated machines.
- Creation of a pipeline monitoring tool and event-driven + serverless blue/green deployment.
- Implementation of E2E test framework in code merge pipeline.
- Addition of a dynamic quota management framework on the platform
Failures
- Incomplete monitoring during certain regressions (insufficient healthchecks in certain scenarios).
Experience feedback
My most accomplished experience: I learned the rigor of software development at scale and operational rigor. If Medicalib had allowed me to understand the difference between a small startup and a 30-40 person structure with clients, Mindflow made me transition to a much more demanding approach — precision, observability, and compliance.
Many excellent advances over these 3 years, in terms of team processes. On the IT security side, delivery, QA processes, development standards, among others. Really pleasant to find yourself with a team that listens, and ready to disagree in commit when it’s necessary for the good of the project.
If I have to remember a bit of negative, in 2 years, without having recruited or signed many clients, we bet on features that we “estimate necessary because the competition does it” or that “we think very good for our clients” without knowing if it’s exactly what our clients want. We therefore spend a lot of time making very rigorous product processes, of features that are not necessarily adapted to our market. When n8n does something, it doesn’t mean that our clientele wants it, they don’t have the same budget and the same team, nor the same clients. It’s more important to remain adaptable and dynamic, do good studies, and iterate quickly with small deliveries to get field feedback quickly, whether from our clients or the market.
Thank you for reading!