Builds
Work that runs.
Four systems we designed, trained and put to work. Each written up the same way: the problem, the build, how it runs.
- 01CatalystSales and marketingLive↓
- 02Predictive MaintenanceIndustrial motors and machinesDeployed↓
- 03Weld Defect AnalysisNon destructive testingDeployed↓
- 04Operations AuditLogistics and manufacturingPartnering↓
Catalyst
Sales and marketingLive
A sales team made of tuned small models.
- Problem
- Our services solve different problems for every company, and often for every team inside it. Reaching the right person meant researching the account, the teams and what each one stood to gain. Generic AI writers only changed the first name.
- Build
- A chain of fine tuned small language models, each trained for one job. Together they research the account, map the problems we solve for each team, put a number on what those problems cost, and run the conversation in phases until there is a decision.
- 01
Research
The company, its teams, how it actually operates.
- 02
Map
Which problems we solve, and for whom.
- 03
Cost
Cashflow gained or burn avoided, in their numbers.
- 04
Converse
Replies, objections and follow ups, in context.
- 05
Act
Book the meeting, hand over, or step away.
- Models
- Fine tuned SLMs, one per task
- Actions
- Meetings, handoffs, objections, removals
- Runs on
- Your infrastructure, no per token bills
- Available
- Custom builds for each industry
Predictive Maintenance
Industrial motors and machinesDeployed
Every machine announces its failure. Long before it stops.
- Problem
- Unplanned downtime is the most expensive way to learn a bearing was worn. Calendar maintenance replaces parts that had years left and misses the ones that did not.
- Build
- Deep learning models trained on vibration, motor current, temperature and acoustic signals from each asset. They learn what normal looks like for that machine at every load, flag the drift, name the likely fault and estimate the remaining useful life.
- 01
Sense
Vibration, current, temperature and sound at the bearings.
- 02
Learn
A baseline for every asset, at every load.
- 03
Detect
Drift from normal, well before an alarm would trip.
- 04
Diagnose
Bearing, imbalance, misalignment, looseness, winding.
- 05
Plan
Remaining useful life, inside the next maintenance window.
- Signals
- Vibration, current, thermal, acoustic
- Output
- Fault class and remaining useful life
- Runs on
- Edge nodes beside the machine
- Connects to
- Your CMMS, as work orders
Weld Defect Analysis
Non destructive testingDeployed
Every weld read like your best inspector reads it. Every time.
- Problem
- Ultrasonic, magnetic particle and radiographic data is still read by hand, one scan at a time. Calls vary between inspectors and shifts, and the backlog holds production at the gate.
- Build
- Deep learning models trained across UT, MT and RT data. They locate each indication, classify the defect, size it and check it against your acceptance criteria, then hand the inspector a ranked queue to confirm.
- 01
Ingest
Phased array scans, MT images, films and digital radiographs.
- 02
Locate
Every indication found and outlined.
- 03
Classify
Porosity, slag, lack of fusion, cracks, undercut.
- 04
Size
Length and depth, judged against your criteria.
- 05
Review
The inspector signs off. The model learns from it.
- Modalities
- UT, MT and RT in one model family
- Output
- Location, class, size, verdict
- Sign off
- A qualified inspector, on every call
- Runs on
- Your network. Scans never leave site
Operations Audit
Logistics and manufacturingPartnering
We find where the operation bleeds. Then we stay and fix it.
- Problem
- Margin leaks through a hundred small places: an idle machine, a half empty truck, a system nobody trusts, a step done twice. Reports name them. Few teams have the hands to close them.
- Build
- An end to end audit of hardware, operations and software, run on the floor with your people. We cost every bottleneck, design the fix and deploy it ourselves, with AI, operations research, supply chain optimisation and network design where they pay back.
- 01
Walk
Every line, dock, lane and system, on site.
- 02
Measure
What each bottleneck and bleed costs per year.
- 03
Design
Hardware, process and software fixes, ranked by return.
- 04
Deploy
Our consultants, working inside your organisation.
- 05
Hand over
Running systems and a team that owns them.
- Scope
- Hardware, operations, software
- Methods
- AI, operations research, network design
- Delivery
- Our consultants, on your site
- Terms
- A partnership, not a report
Your problem could be build 05.
Bring us the process that costs you most. We scope it, build it and stay until it runs.
