Ohmni
A schematic says 2.50 V. The meter reads 0.45 V. I built Ohmni to help explain that gap: confirm the circuit, calculate what it should do, and choose the next measurement that separates plausible faults.
Origin: BuilderBase Multi Model Hackathon · SF Tech Week · October 11, 2026. I built the prototype and continued developing the diagnosis and verification workflows after submission.
01 / THE PROBLEM
The drawing and the measurement disagree.
For a 10 kΩ / 10 kΩ divider on a 5 V supply, the expected midpoint is 2.50 V. A measured 0.45 V is evidence to investigate. Ohmni tests alternate component values against the observation; in the recorded demo, a 1 kΩ lower resistor ranks as the first hypothesis.
The product connects tasks that usually live in separate places: reading the schematic, checking connectivity, solving the circuit, recording a meter reading, and deciding what to probe next.
02 / WHAT I BUILT
Model-assisted reading. Solver-owned calculations.
I also added provider fallback behavior, request limits, server-side validation, and traces that show failures and skipped calls alongside successful ones.
03 / EVIDENCE & LIMITS
A diagnosis is a testable hypothesis.
The repository includes the divider demonstration, solver tests, synthetic schematic fixtures, and recorded UI flows. The hackathon diagnosis path covers DC resistive circuits; its single-fault ranking is relative to the faults it models. Real-photo extraction accuracy and time saved have not been measured.
The app rejects malformed or unsolvable graphs and can report that no modeled fault fits. Two models agreeing on an image does not remove the need to confirm its connections.