AGENT
Tron
Persistent local operator agent on DGX. Day-to-day automation, research, and build loops without cloud tax.
800+
Enterprise AI users (TAVA)
neuralyogi.com · operator HQ · personal surface
I’m Marc Mailloux — Senior Principal AI Engineer and Chief Engineer for sector-wide AI at Northrop Grumman (TAVA). This site is the command surface for what I build: secure LLM/RAG platforms, autonomous agent infrastructure on DGX hardware, and consulting that ships systems — not slide decks.
Pilot-tested local multi-cam notes: quiet alerts, IP/PoE default, USB as special case, Sense craft path. Interest waitlist open — no charge yet. Full pack is a paid zip after checkout, not a free dump.
Most “AI portfolios” stop at demos. Neural Yogi is the opposite: patterns that survive secure enterprise constraints and the messy reality of agents that run unattended on local metal.
By day: hybrid RAG, MCP tool servers, domain assistants, and platform ownership for hundreds of internal users. Off-hours: fleets of coding agents (Loopie), research simulators (Dark Factory), fleet command (NERV), and a paired LLM ops stack on DGX Spark.
Open WebUI pipelines, hybrid retrieval, domain assistants, secure deployment, MCP-backed tools over internal data. Platform ownership for scale (TAVA: 800+ users).
Vision-driven agents, self-maintaining roadmaps, babysitter supervision for stalls, fleet APIs and notifications. One agent per project that actually ships.
Paired models (heavy coder + light general), LiteLLM routing, concurrency for multi-agent sessions on DGX unified memory — zero marginal inference cost for personal ops.
Persistent research agents, graph/wiki memory, scenario compilers, Monte Carlo worlds — compare many futures instead of one chat answer.
Model Context Protocol servers, containerized training/deploy, CI/CD for AI services, eval-minded handoffs teams can run without the original author.
Persistent local operator agent on DGX. Day-to-day automation, research, and build loops without cloud tax.
Unified-memory local inference: MoE coder for deep work, light model for fast tasks, concurrent agent sessions.
Single OpenAI-compatible gateway. Aliases for heavy-coder / light-general so tools and agents stay portable.
Fleet status, stall detection, GPU/LLM activity — command layer for multi-agent production work.
Production fleets of local software engineering agents. Vision north stars, self-updating roadmaps, babysitter supervision, fleet monitor + API. Walk away — it ships.
Agentic research with durable graph/wiki memory and OASIS Monte Carlo simulations for live market and narrative theses.
Sector-wide internal AI platform — Open WebUI + Azure OpenAI, hybrid RAG, MCP agents, domain assistants. Chief Engineer ownership.
Paired Qwen MoE coder + light general via LiteLLM. Tuned concurrency for multi-agent + IDE sessions on unified memory.
Ops console for agent swarms: live status, stalled detection, GPU/LLM activity, dashboards for Loopie-scale work.
Static NERV HQ site — resume, work archive, consulting. nginx-only Docker on DGX; same compose pattern for a future VPS.
Map stack, risks, and highest-leverage AI use cases. ADRs and a build/buy roadmap you can defend to leadership.
Ship a working agent, RAG, or platform slice in weeks — with deploy path, docs, and handoff your team can own.
Ongoing architecture, new workflows, enablement. Priority async access when production AI is a continuous program.
Full resume (ATS PDF + on-site view). GitHub for public craft. Work page for system narratives behind LOCAL builds.
Hiring managers: start with the resume. Collaborators and consulting leads: open CONSULT or reach out on LinkedIn. This is the only public surface you need — neuralyogi.com.
LinkedIn marcemailloux@gmail.com Download PDF