.png&w=256&q=75)
Marcus Comotti@mcomotti
4
Events attended
2
Submissions made
Brazil
20+ years of experience
About me
Multidisciplinary technology entrepreneur, researcher, and systems architect with over 20 years of international experience across IT infrastructure, engineering, project management, and applied research. Founder and developer of JARVIS / PROMETHEUS X6, a distributed AI orchestration and governance architecture designed to connect intelligent agents, heterogeneous computing resources, enterprise systems, and real-world devices. My current work focuses on agentic AI, secure execution, AI governance, distributed computing, GPU acceleration (AMD ROCm and NVIDIA CUDA), and the integration of open-source technologies into practical, auditable systems. With an academic background in archaeology and archaeometry, complemented by postgraduate studies in artificial intelligence and an MBA in project management, I bring an interdisciplinary perspective to complex technical challenges. I believe the future of AI lies not only in more powerful models, but in building trustworthy systems capable of reasoning, coordinating, and executing responsibly across physical and digital environments.
🤓 Latest Submissions
PROMETHEUS FORGE
PROMETHEUS FORGE helps maintenance technicians understand machine signals through voice, while keeping execution authority and decisions under human control. The working prototype uses a physical ESP32-C3 with an OLED display to emulate pump P-204 in NORMAL, WARNING and CRITICAL scenarios. It publishes explicitly simulated telemetry over authenticated TLS MQTT. Local rules establish severity. A spoken status query reads telemetry without GPU inference; an explicit analysis request reaches Qwen3-32B on one AMD Instinct MI300X through the existing JARVIS Core, PROMETHEUS X6 runtime and AMD executor. The existing Xiaozhi and AssemblyAI bridge carries spoken requests and responses. Strict contracts constrain model output, preserve request correlation and prevent automatic execution replay. No physical actuation is authorized or performed. When newer telemetry supersedes the analyzed sample, the response is marked historical and current recommendations are suppressed. Separate WARNING and CRITICAL backend runs passed validation, with recorded service intervals of 6.128 and 8.466 seconds. These are individual observations, not benchmark guarantees. The English demo combines synchronized desktop and physical-device footage. Evidence distinguishes operator listening confirmation from an instrumented playback acknowledgement. FORGE contributes industrial contracts, edge firmware, telemetry ingestion, governance and evidence capture. Core/X6 and voice transport are preexisting dependencies. The public repository documents this boundary and provides independent contract tests. The intended business benefit is faster, clearer incident investigation. A future plant pilot would measure operator comprehension and investigation time; reduced downtime or certified industrial safety is not claimed.
18 Oct 2026
PROMETHEUS VOICE
PROMETHEUS VOICE connects English speech to bounded tools and physical devices while keeping users in control. AssemblyAI powers voice interaction through a browser or a Zhengchen ESP32-S3 device. The Voice application handles tool handoffs and connects to PROMETHEUS X6 through its WebSocket interface. A camera request triggers a permission question. Only a fresh spoken confirmation authorizes one photo, and the detector returns a prediction to the same conversation. Sentinel supplies filtered aircraft display counts and weather observations. ShadowBroker supplies attributed headlines, and the X6 registry exposes registered capabilities without claiming live availability. A separate local arithmetic workflow lets users review a plan before execution and inspect the result and its hash-linked evidence receipt. The recorded demonstration shows camera consent, Sentinel queries and a verified local calculation. The initial audience is builders and operators who want voice access to connected systems with understandable controls and visible results. The contribution is the Voice layer, confirmation experience and tool integrations. X6, device firmware and external services are pre-existing components. General distributed execution from Voice remains planned. The public repository includes the Voice source, tests, setup instructions and integration contracts. The application runs locally; voice requires AssemblyAI access, and hardware workflows require compatible external services. Detection predictions can be wrong, and telemetry has source-specific freshness limits.
30 Sep 2026
