University / research collaboration · October 18, 2026

Research with
Kira Labs.

I'm Robert McMurrer, an independent researcher developing NewBrain, synthetic-person experiences and experimental virtual bodies. I am seeking faculty and student collaborators who can help turn clearly scoped prototypes and negative results into reproducible studies.

No university sponsorship, affiliation, research approval, equipment award or funding is claimed.

Research opportunities

Three connected questions.
Separate evidence.

Each proposed collaboration would begin with clear boundaries, small test designs, and independent review.

Machine learning

Can NewBrain learn complete answers?

Investigate optimization, token generation, stopping behavior, and retention without adjusting consumed Original12 scoring. Compare carefully matched new experiments with an independently qualified baseline.

Public2: 0/2 → 1/2 after repeated training and reload; separate Original12: 2/12 baseline → 0/12 trained.

A separate completed read-only diagnostic has now passed all four of its scoped execution controls; its combined loss matched token-weighted row losses, while the three-token response still ended immediately. This narrows one accounting question without establishing a learning fix.

NewBrain research →
Embodiment

Can measured body events become useful signals?

Reproduce 512-sampled-vertex pose/restoration, inspect the separate 8/8 synthetic skin-event component, then qualify each interface before attempting a brain–body link.

No physiological lung, real neural input, calibrated skin or closed-loop NewBrain control has been demonstrated.

Avatar and Kira World work →
Human-centered research

Can personalized companions help learners?

Design consent-based studies of conversational tutoring, historical-persona exploration, lesson breaks, accessibility and digital continuity. Begin with educational content rather than clinical or therapeutic claims.

The Temporary AI Creator has draft identity profiles and constrained examples, not proven learning outcomes.

Published walkthroughs →

Small first collaboration

A proposed 30–90-day pilot.

A limited project, not a promise of results or a request for open-ended laboratory access.

Days 1–30

Reproduce and profile

Agree on a sponsor, access terms, licenses and a bounded protocol; qualify a public control, record baseline CPU/RAM/VRAM/time and retain failed runs.

Days 31–60

Investigate one cause

Run a predeclared learning-versus-generation experiment, preserving original test questions and avoiding post-hoc score repair.

Days 61–90

Review before expanding

Evaluate memory/retrieval or body-event tests with domain specialists. Decide whether a next experiment or hardware allocation is justified.

What would be most useful

First: one faculty sponsor, one student research partner (or approved graduate collaborator), regular mentoring, secure access to shared lab/research computing and a reproducibility checklist.

Later if justified: shared GPU/HPC time, RAM or GPU-laptop loans, approved after-hours access, VR hardware and measured sensor equipment. Current RAM/VRAM/storage targets are planning options—not minimum requirements, and equipment alone cannot solve a learning defect.

Any human-participant or patient-record research would need separate institutional approval, consent, appropriate privacy controls and domain review. Begin with approved public and synthetic data.

Materials to inspect

Public first. Private by permission.

The detailed NewBrain, KiraWorld, Patient Horizon and Aster development repositories are private. Public readers should use these source-safe summaries; original files, technical methods and project-specific access can be discussed with the owner under appropriate permissions.