How it works
A walkthrough of one incident from start to finish showing how Upstrix guides the repair, captures what worked, and makes the next diagnosis sharper.
What's inside Upstrix
None of them asks your technician to find the right words.
DIAGNOSIS ENGINE
No chatting. Describe what you see, and a statistical model computes the most likely causes along with the evidence behind each one.
KNOWLEDGE DATABASE
Not a folder of PDFs. A structured model of your equipment including machine types, named root causes, symptoms with diagnostic weights, and repair steps reviewed and approved by your seniors.
AI ASSISTANTS
LLMs handle messy human input well by normalizing voice and free text into structured observations. They don't decide what's wrong. The statistical model does that.
Why it gets better over time
One initial setup, then a cycle that runs on every incident.
Month one, it reflects your documents. Month six, it reflects six months of actual repairs on your machines. These are reviewed, structured, and immediately available to whoever opens the next incident on that equipment.
Initial setup
Upload whatever you have, such as OEM manuals, maintenance logs, technician notes, and Excel cookbooks. Upstrix drafts the initial knowledge structure: candidate machines, root causes, symptoms, repair steps. Your seniors review and correct.
Ready for real incidents within days. No CMMS connector, no telemetry, no IT project required.
MAKE THE SYSTEM ALIVE, DAY ONE
Spec sheets, procedures
Past incidents, fixes, preventive schedules
How your team actually does it
Of your machines, your line, your setup
What your experts already explained out loud
Most maintenance AI projects stall because they require sensor installations or ERP integrations before any technician sees value. Upstrix works from uploaded documents and human-reported symptoms alone. Integrations come later as upgrades rather than prerequisites.
See the incident
Before the technician touches the machine, Upstrix has already identified the most likely causes based on what the operator reported, the machine's history, and every previous incident on that equipment type.
THE MACHINE AND ITS HISTORY
One tap shows full incident history for this machine and its type. Recurring failures are visible immediately, and the cause ranking already reflects the pattern.
WHAT'S ALREADY BEEN REPORTED
Every symptom is tagged by source: operator on the line, technician input, or machine telemetry when connected. The technician sees the situation as it actually arrived.
THE MOST LIKELY CAUSES
Each cause shows a likelihood percentage and the specific symptoms behind it: confirmed, conflicting, or unverified. The technician sees not just what the system thinks, but exactly why.
Sharpen the diagnosis
The technician speaks or types freely. The system extracts each distinct observation, maps it to known symptoms, and recomputes the most likely causes on the spot.
When uncertainty is high, Upstrix asks targeted yes/no questions to narrow the field. Each answer updates the result in real time.
Repair guidance, built by your team
Every procedure in Upstrix has been reviewed and approved by a senior on your team. Not generated from a manual. Authored, verified, and attached to the exact root cause it belongs to.
WHERE YOU ARE IN THE REPAIR
Previous step, current step, and next step are always visible. If the repair changes hands, the next technician picks up exactly where the previous one stopped.
INSTRUCTIONS THAT MATCH YOUR MACHINES
Photos from your line, voice notes from your seniors with name and date, and links to the right page of the right manual. This is not a generic procedure, but rather the way your team actually does it.
WHEN THE STEP DOESN'T WORK
The technician flags the step, describes what they're seeing, and the model determines whether this is a mis-ranked cause or a new failure mode. Escalation carries the full activity history, ensuring the next person doesn't start from zero.
Capture what worked
After confirming the fix, the technician gets one prompt: anything worth adding? A 10-second voice note is enough. Seniors review every proposal before it goes live. Nothing reaches the knowledge base without approval.
KNOWLEDGE ASSISTANT EXAMPLE
Senior says (30 seconds):
"For Line A use a Torx T30, not the 5mm Allen key. And before you remove the blockage, rotate the shaft a quarter turn backwards, otherwise the bearing seizes."
Upstrix proposes:
→ Replace "5mm Allen key" with "Torx T30 driver" in Step 2
→ Add warning to Step 3: "Rotate shaft quarter-turn before removal to prevent bearing seizure"
Senior clicks Accept. Live in 10 seconds.
Learning loop
Two ways the team gets better.
WHEN A NEW SYMPTOM IS ADDED
The system learns the new signal. Next time it appears on this machine type, it counts toward the most likely causes from the start for every technician.
WHEN A REPAIR STEP IS UPDATED
When a senior approves a change, it propagates immediately. Every technician who opens that root cause from now on sees the updated version.
The shift, at a glance
The coordinator sees active incidents, who is assigned, and what step they're at. The maintenance manager sees the same data in summary across shifts and days.
When something escalates, the full incident history travels with it. Decisions are made on evidence.
Why we built it this way
01
Most maintenance AI puts a chatbot between the technician and the answer. We think that's the wrong design for someone standing next to a broken machine. So we removed the conversation entirely.
02
Most maintenance AI projects stall waiting for sensor installations and ERP integrations. We decided no technician should have to wait for IT before seeing value.
03
Your ERP tracks costs. Your CMMS tracks work orders. Neither tells a technician what is most likely wrong with Line 3 at 2AM.
We're selecting 3-5 manufacturing companies as Design Partners. You shape the product. You use it free, forever.
Not ready to commit? Join the waitlist, and we'll reach out when the next cohort opens.
