Print-friendly · Prepared October 18, 2026
Kira Labs
Research Brief
Independent applied AI research by Robert McMurrer, based in Newark, New Jersey.
Research purpose
Explore reusable learning architecture (NewBrain), synthetic individuals with distinct identities, memory, voice, and virtual environments (Kira World), and clinically reviewed research possibilities through separate Patient Horizon and robotics projects. The long-term aim is useful continuity and accessible experiences; these benefits are not claimed as demonstrated.
Current measured evidence
| Track | Result | Interpretation |
|---|---|---|
| Public2 (trained-on) | 0/2 → 1/2 | Short answer acquired after 64 updates and saved-state reload; longer answer failed |
| Original12 (separate) | 2/12 → 0/12 | Regression; two baseline passes represented absent-information controls |
| Numerical gradient sample | 30/30 at three step sizes | Independent internal saved-data review; local sample only, not all derivatives |
| Objective/generation (read-only) | 4/4 execution controls | Three target-token losses: 1.3116488, 2.4736416, 2.0863106 combined; long answer still immediate EOS |
| Avatar geometry | 512 vertices / 5 poses | Exact commanded restoration, not realistic passive tissue |
| Synthetic event component | 8/8 checks | Software ownership/reset controls, not working biological nerves |
Application areas to investigate
Student-paced educational companions and source-grounded historical personas; accessibility and companionship; synthetic personal history and authorized digital twins; virtual and future robotic embodiments; and clinician-reviewed patient-record research. No demonstrated clinical, therapeutic or classroom effectiveness is asserted.
First collaboration request
A faculty sponsor, one student or graduate research partner, weekly mentoring, approved research-computing/lab access, reproducibility review, and a bounded 30–90-day pilot. Shared resources and equipment loans are preferred before new purchases. Institutional permission, data rights, and safeguards would be agreed before tests.
Selected public references
Research evidence dashboard · Dated milestones · Published demonstrations · Methods and limits
Evidence cutoff: October 18, 2026. Internal checks are not external academic replication. Source files and private repositories are shared only through an approved access process. No university affiliation is implied. Contact: rmcmurrer@kiralabs.org · Business: (317) 586-8199.