Brain-response analysis.
Personal AI.
Built where Cardinal meets Code.

What we ship

Three doors into the lab. Two are open-source research projects; the third is the storefront where ABM LLC ships handmade artifacts of that research.

What ABM does

Alexios Bluff Mara LLC partners with the ISU College of Applied Science & Technology on applied AI and computational-neuroscience research only (not coursework). The work is hardware-flexible by design: the same code path runs on a single RTX 5090 in our Chicago lab (call sign Ascended Base), on the M4 Max MacBook (Big Apple) sitting next to it, and — through ACCESS-CI partnerships at UIUC and similar institutions — on national HPC nodes when the experiment demands them. Open-source where possible. Local-first by default.

Active named collaborators at ISU CAST: Mangolika Bhattacharya, Sally Xie, and Rosangela Follmann (early-stage, one meeting). Full bios on /about.

Below are the ISU departments and facilities the lab interacts with. None of these endorse the lab's work; they are listed because they are the academic homes the research draws from and contributes to.

Active hackathon submissions

Both projects are competing in May 2026 hackathons. Submissions are live and publicly reproducible.

Receipts — the real numbers

Pulled from the live project READMEs. No extrapolation, no cloud bills rounded up.

Work with us

Alexios Bluff Mara LLC is actively seeking research partners, compute sponsors, and early investors aligned with local-first AI infrastructure and computational neuroscience.

Funding & capital

What Mercury says when you ask it who it is

The output below is from a real Mercury terminal session running on Seratonin (the local 5090). Same agent answers identically through the Discord bot (@abmsnowy) and the phone-friendly web dashboard.

$ mercury chat
[mercury] connected to local Ollama (gemma4:e4b @ 194 tok/s)
[mercury] session resumed · 1,247 prior messages in context

› who are you?

Identity & Architecture
  Agent:                Snowy The Bot / ABM Hermes
  Developer:            Alexios Bluff Mara LLC / Red Team Kitchen
                        in collaboration with Illinois State University 🔴
  Execution Environment: Local RTX 5090 via Ollama (direct, low-latency,
                        high-power access to GPU resources)
  Model:                Gemma 4 E4B (194 tok/s)
  Cost / Limits:        No cloud API costs, no rate limits
  Persistence:          Persistent memory across sessions

Software Capabilities (Skills & Tools)
  Web/Browser:    web_search, web_extract, browser suite (Navigation, Vision, Interaction)
  Code/Dev:       terminal, read_file, search_files, patch, write_file, github,
                  systematic-debugging, test-driven-development
  Data/MLOps:     data-science, mlops (TRL/Unsloth fine-tuning, vLLM inference,
                  SAM segmentation), research (arXiv, literature review)
  Memory/Workflow: memory (user/general facts), skills_list, session_search,
                   delegate_task for orchestrating multi-step workflows

Conclusion
  Locally hosted, powerful, state-of-the-art AI agent designed for
  uninterrupted, resource-intensive, multi-domain work. Primary vulnerability
  surface: inability to run standard system diagnostic commands in the
  immediate shell environment. All operational parameters (GPU, memory, model
  throughput) are confirmed internally.

What we have, and what we'd build next with funding

The next surface for both Cortex and Mercury is spatial — head-mounted displays where a brain-response visualization actually wraps around the viewer instead of being trapped in a browser tab. Three platforms matter here, and we already own two of them.

Cost-to-ship math: Quest 3 + AVP combined puts the lab at ~$4,000 of XR hardware on hand. Adding Galaxy XR + Google Glasses would be another ~$3,000-$3,800. Total spatial-computing capex to cover all four major platforms: under $8,000. That's the whole BOM for an education-sector spatial-neuroscience pilot. Compare to the cost of one Hospital MRI hour ($600-$1,200) and the conversation gets short.