Cyril Moron

Technical owner of the shared software platform (framework) behind Naval Group's combat management systems (CMS): 45 components for distributed systems. Tech lead for a team of 2 in-house developers and 3 to 4 contractors. Keeping fielded systems running while modernising core services and engineering practices.

• Design and maintenance of the CMS distributed middleware: real-time data exchange (DDS/DLRL), distributed configuration (JGroups), machine and process fault tolerance, remote command over CORBA.
• Rolled out agentic AI to the team as the bridge with internal AI R&D: OpenCode and Hermes agent on self-hosted models in a private cloud, RAG and an agent harness for framework development and delivery.
• Agentic development loop that builds and tests in Distrobox on the in-house legacy toolchain (Ant-based), domain skills for internal technologies, and docs as code: legacy documentation migrated to Markdown versioned with the code, feeding the agents.
• Final escalation point for incidents on fielded systems: network latency, distributed data corruption, JVM memory leaks, system crashes.
• Migrated the CI/CD pipeline from Jenkins to GitLab CI and Nexus. Own release management, versioning and artefact integrity.
• Own the company-wide Python standards: coding guidelines, tooling, packaging. Unit tests (JUnit, pytest), black-box tests, TDD.

Software integration and system configuration (OS, network, startup sequences) for submarine CMS, mainly on the Scorpène class (Brazilian PROSUB programme).

• Integration and validation of the surveillance and weapon subsystems: sensor data fusion and classification, a shared tactical picture identical on every console.
• System performance analysis and tuning: scalability, boot time, availability.
• Technology transfer: trained and supported the Brazilian Navy until it was autonomous on CMS assembly.

Development and integration for the Barracuda and Scorpène combat systems: weapon systems, hardware interfaces, tactical situation.

• Designed and implemented the weapon launch sequence state machines in Java, described in XML: SM39 (Exocet), F21 (torpedo), MdCN (naval cruise missile).
• Implemented supplier network protocols from their ICDs (MdCN).

Technical exploration on an EVM L2: a team of AI agents picks the leads, writes the code, deploys and measures. The human sets the risk limits and signs off every commitment.

• Smart contract auditing (Solidity, DeFi) by an agent pipeline, validated by replaying already-judged contests (12 of 23 High/Medium findings recovered, against 8 for a single agent), then entered in a Cantina contest.
• Custom agent harness: orchestrator and specialised sub-agents (development, adversarial review, verification), versioned memory, automatic session recovery, 24/7 operation with a human in the loop. Guardrails: agents never see a key, isolated and capped signer.
• On-chain tooling in Rust and typed Python: real-time sequencer feed decoding, exact simulation of Uniswap pools, about 3,600 tests.

Public site (pace calculator, search over 940,000 French Athletics Federation athletes, weekly scraper) and a MyPacer Club module that sends club officials the weekend results every Monday. FastAPI, SvelteKit and PostgreSQL, deployed with Docker Compose and GitHub Actions.

• Prospecting, follow-ups and season renewals delegated to Hermes agent through a Python CLI designed for an LLM: JSON output, local and remote modes, scoped tokens, Markdown operator skills loaded on demand.

Multi-model routing: a frontier model orchestrates and delegates to lighter models based on capability, latency and cost. Dedicated KVM VM, secrets kept out of the workspace, idempotent Ansible deployment.