Building AgentForge QA, a private personal AI runtime, and AYYO — a humanoid robot in development. The machines are supervised. Mostly.
DEPLOYING SOMEWHERE BETWEEN
CONFIDENCE AND CHAOS.
machine learning / agents / algorithms / code
I'm just making sure
they don't get too confident.
Move the cursor.
The lab gets nervous.
Static diagnostics.
Touch mode active.
Static diagnostics.
Reduced motion active.
Primary systems / 02
Verification. Personal intelligence. Humanoid embodiment. Each world moves only when it has a reason.
01 / primary system
Coding agents make claims. AgentForge QA checks those claims against repository evidence before the work is accepted or pushed.
Built a conservative CLI that inspects configured commands, required artifacts, claims, Git state, and risk before returning an evidence-backed verdict.
Explore proof ->Verification chamber / evidence scanner
CLAIM → EVIDENCE → RISK → VERDICT
02 / primary system
An owner-controlled personal AI runtime connecting models, tools, context, and everyday computer workflows through one coherent workspace.
Building an evolving private assistant environment around deliberate owner control, inspectable work, and practical day-to-day use.
Private by design.Private control room
MODELS + TOOLS + CONTEXT → ONE CONTROLLED WORKSPACE
03 / primary system
A humanoid robot in development, advancing through simulation before physical hardware claims are made.
Co-building the software foundation across perception, intelligence, manipulation, and embodiment. Current public progress includes simulation-only grasp interaction; physical validation remains explicitly unclaimed.
Explore proof ->Robotics / embodiment lab
SIMULATION → MANIPULATION → EMBODIMENT / HARDWARE: NOT CLAIMED
Evidence ledger / 03
Systems, robotics, experiments, and fixes. Every public claim points to inspectable evidence.
Current signal
CLAIM → EVIDENCE → RISK → VERDICT
Build-night archive / 04
Not a trophy shelf. A pressure test for deciding what matters, proving what works, and cutting what does not.
01 / isolate signal
02 / test constraint
03 / save proof
Console B / awake too long
Local-first builder runtime
Planning, tool use, approval, generated artifacts, and proof should stay visible in one local workflow.
Rover perception experiment
Terrain and obstacle segmentation for autonomous off-road rover perception.
Autonomous racing pipeline
A monocular visual-servoing pipeline that closes the loop from camera observation to ROS 2 control interfaces.
Business systems experiment
Turn small-business activity into a structured operating picture and prioritized decisions.
FAILURE IS AN INSTRUMENT.
OBSERVE THE BREAK.
FIND THE ASSUMPTION.
REBUILD WITH EVIDENCE.
Human process / 04
A professional pull-request opener. And apparently, still not done.
Capability atlas / 05
Deterministic evidence, conservative verdicts, and visible risk before trust.
A good problem is a perfectly valid introduction
Good. Those are usually the interesting ones.