Manufacturing test

Agents that runthe test line.

FIXT ingests the schematic, datasheets, and firmware, generates the station test plan, runs and monitors it on the line, and triages every failure.

fixt run --fixture bench-04 --dut WX-2.3-revC
PWR-01rail_sequencing412 ms
PWR-04inrush_current233 ms
CLK-02xtal_startup189 ms
IO-11i2c_enumerate96 ms
RF-03ble_tx_power780 ms
OPT-07ppg_led_uniformity1140 ms
triage · RF-03

Fixture, not silicon. The same −0.1 dBm offset appears on every unit tested at bench-04 since the cable swap on rev C. Units on bench-02 pass at 4.1 dBm. Recommend re-running path loss calibration before touching the board.

confidence 0.9134 comparable runs2 similar incidents
01The problem

Manufacturing test still needs a person watching it.

Someone writes the station scripts, keeps the line running as revisions ship, and reads logs by hand when a unit fails — was it the board, the fixture, or the test itself?

Root-causing usually means pulling in failure analysis, reliability, ATE, and quality, each holding a different slice of the picture, and getting them to compare notes.

That doesn't scale with volume, and it doesn't scale with how often hardware changes.

02What we build

Full test-line coverage in days instead of weeks.

01

Ingest

The schematic, the datasheets behind every part, and the firmware that runs on it.

02

Generate

The station test plan, with limits derived from the parts themselves.

03

Run

The line runs and stays monitored, through every revision that ships.

04

Triage

Each failure read with the cross-domain view four engineers would bring.

03Positioning

An operator, not a copilot.

Most AI for hardware stops at helping you write it faster. FIXT runs the station, watches it, and knows when something is wrong — closer to a shift you don't have to staff than a tool you use.

First-pass yield · line 2 · WX-2.3 rev C

last 10 shifts

Fixture drift caught on shift 4; recalibrated after shift 6.

8085909510096.4%Shift 1Shift 5Shift 10
04Who we are

One of us knows what a test engineer needs. The other has shipped agents under real constraints.

Akkhil

Built distributed test infrastructure, automated optical validation, and embedded diagnostic systems at WHOOP. Ran electromechanical validation and quality systems automation at Avanos.

Shash

Ships production agentic AI at Bridge, where multi-agent systems and retrieval over messy real-world documents have moved over $500M in financing.

We're looking fordesign partners.

One station, one fixture. Schematic in, a running and monitored test cycle out.