Independent test systems consultant
Automated product test, built in Python, delivered on a fixed price.
Benchtop testers, ATE racks, and robot-tended inspection cells for manufacturers between Springfield and Hartford. No seat licenses, no vendor lock, and a firm number before the work starts.
Replies within 24 hours. Currently taking select projects.
- Cycle
- 30.6s
- Manual
- 415s
- Result
- ...
ST-01 / DUT-4417-0026
Example sequence, not a client result. Timings are what these measurements cost on a station.
- Keyence certified
- Machine visionVision is the payload in most tended cells, so the certification is the one that matters.
- 5 years
- Building test systemsBenchtop testers, ATE racks, and the software that runs them.
- 24 hours
- Response, stated upfrontWeekly written status beats a standing meeting. Calls at lunch or after five.
- 1 hour
- From South HadleySpringfield to Hartford, plus the Connecticut and Vermont edges. I can be on your floor this week.
Instruments and robots the platform already drives
- NIPXI and DAQ
- KeysightInstruments
- PickeringSwitching
- KeyenceVision
- CognexVision
- Universal RobotsCobots
- FANUCRobots
- RoboDKSimulation
Where this usually starts
Three versions of the same afternoon
Every conversation about test automation starts with one of these. Usually two of them at once.
Test is the reason the ship date moved
Units stack up behind one bench and one technician. Throughput is set by how fast somebody can read four meters and write the numbers down, and nobody upstream believes that is the real constraint until the quarter closes.
The rack works. The people who understood it left
It was written in LabVIEW and TestStand by someone who is now at a different company. The seat licenses renew every year, the hiring pool for that skill keeps shrinking, and every small change is quoted in weeks because nobody wants to touch it.
Hiring a test engineer takes six months you do not have
The requisition is open. The good candidates want to build products, not maintain a tester. Meanwhile the work that needs doing is a defined piece of software, not a career.
None of these is a headcount problem. All three are an architecture problem, and architecture is a thing you can buy once.
What changes
The same three problems, answered with numbers
60 to 70%
comes from a standard core
The hardware abstraction layer, sequencer, results store, operator interface, and report generator already exist. Your project pays for the edge: your fixtures, your product, your tooling. That ratio is the only reason a fixed price is honest rather than optimistic.
$0
in seat licenses on a deployed station
Python, running on free instrument driver runtimes. Your own engineers can read the sequence, review it, and change a limit without calling anyone. When an instrument goes end of life, one driver changes and the sequence above it does not.
2 weeks
from unscoped backlog to firm quote
A paid assessment produces an architecture, a bill of materials, driver coverage confirmed instrument by instrument, and a fixed price for the build. You can take that document anywhere. What you cannot do any more is guess.
Same twelve checks, both ways
Manual bench
6m 55sper unit
- Wire the unit to four instruments
- Read each meter, write it on the sheet
- Check each reading against the spec by eye
- Type the sheet into a spreadsheet
- File the spreadsheet somewhere
Every number exists once, on paper, in one person's handwriting.
On a Testral station
31sper unit
- Scan the serial, close the fixture
- Sequence runs twelve steps against limits
- Pass or fail decided by the limits file
- Record written with full traceability
- Report generated per unit and per batch
Every number exists forever, queryable, against the serial it came from.
Illustrative, not a client result. The manual figure is twelve measurements read off four instruments and written onto a sheet. The station figure is what those measurements cost when the instruments are driven in sequence.
One platform, two halves
Named parts, not a blank sheet every time
Every job builds the same core and customises the edge. That is what makes the second system cheaper than the first, and the fiftieth cheaper again.
Testral Core
The software every station runs
A test executive, an abstraction layer over the instruments, a results database, an operator interface, and a report generator. Written in Python, deployed with no seat licenses, and readable by the engineers who have to live with it.
- InstrumentsHardware abstraction layer. Every instrument sits behind one interface. Swap a digital multimeter and the sequence above it does not change. This is the layer that stops an instrument going end of life from turning into a rewrite.
- Test executiveSequencer. Steps, limits, branching, retries, and operator prompts. Built on pytest so the test sequence is ordinary Python your engineers can read, review, and run in continuous integration.
- DataResults store. SQLite per station, Postgres plant-wide when you are ready. Serial number, operator, station, firmware revision, and every measured value, on every unit.
- FloorOperator interface. Scan, run, read the result. Built for a technician on a twelve-hour shift, in Qt on the station or in a browser when the station is shared.
- OutputReport generator. HTML and PDF per unit and per batch, driven by limits files rather than by code, so a specification change is a data change.
Testral Cell
The hardware it runs on
Standard rack configurations, and robot-tended test and inspection cells for the loops where a person is loading, testing, and inspecting the same part all shift. Same software underneath, so the cell is an extension of the station rather than a second system to learn.
Cobot
Universal Robots first, FANUC CRX where payload demands it. Loading, tending, and part presentation.
Vision
Keyence or Cognex, for inspection, measurement, and code reading. Certified on Keyence.
Test rack
The same Testral Core stack that runs a benchtop tester, in a rack the cell feeds.
Fixturing and end effector
Designed with a local machine shop rather than a shop I would have to build.
Safety
A documented risk assessment against RIA R15.06 and ISO/TS 15066 in every cell scope. It is a paid deliverable, not a disclaimer.
Simulation
Offline layout and cycle time in RoboDK before anything is bought, so the payback number is checkable.
Where to start
Every price is on this site
Six ways in, all fixed price, all sized so one engineering manager can sign for them without opening a procurement cycle.
Test automation assessment
$3,500 to $5,000
I audit your current test process and hand back an architecture, a bill of materials, and a firm quote for building it.
2 weeks/Remote
What is includedRobotic cell feasibility study
$4,000 to $8,000
Cycle time, payback period, and a simulated cell layout, so the robot decision gets made on numbers instead of a vendor demo.
2 to 3 weeks/Mostly remote
What is includedMigration pilot
$6,000 to $10,000
I port one LabVIEW or TestStand sequence to Python end to end, so you can see the pattern before betting the rack on it.
3 to 4 weeks/Remote
What is includedResults and reporting bolt-on
$4,000 to $8,000
Traceability and reporting added to a rack that already works, without touching the test sequence.
2 to 4 weeks/Remote
What is includedBenchtop test set for one product line
$15,000 to $40,000
A complete functional tester for a single product, built on the standard core, delivered with the operator UI, reports, and traceability already working.
6 to 12 weeks/Remote, then on site
What is includedRemote support retainer
From 5 hours per month
Fixes, tweaks, and operator questions on a system that is already running, billed monthly instead of chased by purchase order.
Monthly, 60 days notice/Remote
What is includedHow the work is staged
Assess, Pilot, Build, Sustain
Each stage ends somewhere you can stop. That is the point of staging it: you buy the next step only once the last one told you something.
- 01
Assess
$3,500 to $5,000/2 weeks
Turn a vague backlog into an architecture, a bill of materials, and a firm number.
You hold a firm quote and can stop here without owing anything further.
- 02
Pilot
$6,000 to $10,000/3 to 4 weeks
Prove the pattern on one sequence before it is applied to the rack.
The approach is proven on your hardware, or it is not, and you found out for ten thousand dollars.
- 03
Build
$15,000 to $40,000/6 to 12 weeks
The full station, delivered against a written acceptance checklist.
Signed acceptance sheet, ninety-day warranty running, final invoice the same day.
- 04
Sustain
From 5 hours per month/Monthly
Someone on the hook when the station stops at 6am.
Sixty days notice, either direction. No lock-in, because lock-in is the thing you are trying to get out of.
Instead of testimonials
Nobody has said anything nice about this yet
There is no case study yet, so instead of asking you to take a stranger at their word, here is exactly how a project closes and what you pay when.
Scope is a checklist, not a paragraph
The statement of work carries a written acceptance checklist. Two pages, agreed before anyone starts, listing what has to be true for the work to be done.
Status arrives weekly, in writing
One email a week: what moved, what is next, what is blocked. No standing meeting, no status call you have to attend to find out where the project is.
Acceptance is walked through on a call
We go down the checklist together, line by line, against the running system. Anything not met gets fixed before the invoice.
Scope changes are quoted, not absorbed
A change order names the work and the price. Nothing gets added silently and then charged, and nothing gets added silently and then quietly dropped.
The platform stays yours to run
You own the deliverables and get a license to the platform underneath. Python on free driver runtimes, no seat licenses to renew, source you can read.
Ninety-day warranty on defects
If something I built is wrong, I fix it. Support after that is a retainer if you want one, and nothing if you do not.
When the first project closes, this section gets replaced by the customer's own words plus two numbers: test time per unit before and after, and weeks from purchase order to running station. Until then, the process is the proof.
Where I work
Close enough to be on your floor this week
South Hadley sits between the Springfield industrial corridor and the Hartford aerospace belt. Integration and commissioning happen in person. Everything else happens remotely, because that is faster and cheaper for you.
- South Hadley, MABaseHome base
- Northampton, MA15 minPhotonics, defense electro-optics
- Chicopee, MA15 minMedical device contract manufacturing
- Springfield, MA15 minPrecision manufacturing, firearms
- East Longmeadow, MA20 minElectromechanical OEM
- Westfield, MA25 minHVAC controls, firearms
- Enfield, CT30 minLaboratory instruments
- Stafford Springs, CT30 minRF and microwave assemblies, PCBs
- East Granby, CT30 minTechnical ceramics, metrology
- Turners Falls, MA35 minWire and cable
- Windsor Locks, CT35 minAerospace systems
- Bloomfield, CT40 minEngine components
- Athol, MA45 minPrecision gauges and instruments
- East Hartford, CT45 minAerospace propulsion
- Simsbury, CT50 minDefense energetics
- Brattleboro, VT50 minOptical filters, aero components
- Worcester, MA55 minAviation audio, medical
- Oxford, MA60 minFiber lasers
Drive times are approximate and measured from South Hadley. Work beyond this radius happens remotely, which for software, drivers, and migrations is most of it.
Asked before the first call
Straight answers, including the unflattering ones
Publishing the gaps is cheaper for both of us than finding them in week six.
How much does test automation cost?
A paid assessment is $3,500 to $5,000 and takes two weeks, and a complete benchtop test set for one product line runs $15,000 to $40,000. Everything in between is published on the services page with a price against it. Fixed bids are estimated hours times rate times a contingency factor, and I track what each job actually took against what I bid it at, which is the only way the next number gets better.
Can you convert LabVIEW or TestStand to Python?
Yes, and the sensible way to start is a migration pilot at $6,000 to $10,000 where I port one sequence end to end and prove the pattern on your hardware. From there TestStand can stay as the executive and call Python modules, so the transition is gradual and nothing gets switched off before its replacement works. Sequences move as they come up for change rather than in one risky cutover.
Do you work on site?
Yes, and being close is the point: South Hadley sits between the Springfield industrial corridor and the Hartford aerospace belt, so most of the target area is inside an hour. Development, driver work, and migrations happen remotely because that is faster and cheaper for you. Integration, commissioning, and factory acceptance happen on your floor.
Are you a firm or one person?
One engineer, taking select projects, which is exactly why the scopes are small and fixed. That is an advantage when your problem is that you need hands on a defined piece of work, and a disadvantage if you need twelve people on site next Monday. If the second one is your situation, say so on the first call and I will tell you.
Are there software licenses to renew?
No. The platform is Python, and the NI instrument drivers run on the free driver runtimes, so a deployed station costs nothing per seat and nothing per year. If your project genuinely requires TestStand, deployment licenses are passed through at cost as a line on the bill of materials rather than marked up.
Where does Python not work for test?
LabVIEW FPGA targets, LabVIEW Real-Time targets such as cRIO, and some of the newest RF instruments still need LabVIEW, and deployment stays on Windows because NI Linux support is partial. This gets verified instrument by instrument during a paid assessment, before a fixed price is signed, rather than discovered afterwards. Publishing the gaps is cheaper for both of us than finding them in week six.
Next step
Tell me what your test bench is holding up.
Thirty minutes, at lunch or after five, on your product and your current test process. No deck. If it is not a fit I will say so on the call, and if I know someone better placed I will tell you who.
Replies within 24 hours. Or write directly to [email protected].