Bring it to us if…
- The problem is physical: light, heat, motion, distance.
- It has to work outdoors, unattended, for weeks.
- You can give us months and a real site to test on.
- You want to own the instrument and the source.
Contract R&D · sensing · optics · field software
We take the problems that need instruments.
Deep Star Labs is a small kept lab for the long haul. We do sensing, optics, and the software that reads them — then we take it into the field and prove it works at 3am in the cold. Engagements run months, not weeks. If your problem fits on a slide, you probably don’t need us.
3–9month engagements
2concurrent programs, max
11instruments on the bench
21:00–04:00night bench window
LOCK 0.7″
TRACKER-04 · drift 0.7 arcsec/min · SNR 42dB
Program burn-down
M4
Month 4 of 7 — low-light photometer for orchard frost.
Bench load
Signal vs. cold
SNR as the instrument drops to −15°C. We chart the failure, then fix it.
What we build
Sensing that survives contact with the world.
Optical sensing
Low-light photometry, spectral discrimination, and lens/filter stacks we spec, mount, and calibrate ourselves.
Sensor fusion
Kalman and factor-graph pipelines that combine optics with IMU, GNSS, and thermal so a single dropout doesn’t blind the system.
Field software
Edge software that runs headless in a pelican case for weeks — logging, watchdogs, and over-the-air patching over a bad link.
Instrument design
Enclosures, mounts, and thermal paths designed for the actual deployment — a roof, a boat, an orchard, not a clean lab.
Calibration & traceability
Every instrument leaves with a calibration certificate and a repeatable procedure your team can rerun.
Night validation
Optics work is night work. We hold a 21:00–04:00 bench window because that’s when the sky and the noise floor cooperate.
How an engagement runs
Months, milestones, and a real field test.
- Site & scope (weeks 1–2)We visit the deployment, measure the light, the vibration, the temperature swing, and write a scope with pass/fail numbers.
- Bench build (months 1–2)Optics stack and first firmware on the bench. We prove the physics before we polish anything.
- Integration (months 2–4)Fusion, enclosure, thermal, and the field software that runs it unattended.
- Field test (a real one)We deploy on your site for at least two weeks of continuous logging — through weather, not a demo afternoon.
- HandoverInstrument, source in your repo, calibration certs, and a runbook. We stay on retainer if you want us on call.
Not a 21-day burn
Two programs at a time
IP is yours
Field data delivered raw + reduced
Retainer optional
Representative programs
What the bench has shipped.
| Program | Domain | Months | Field days | Result |
|---|---|---|---|---|
| Frostwatch | Ag optics | 7 | 21 | −2°C frost call, 40min lead |
| Quayline | Marine LiDAR | 9 | 34 | Dock-edge ranging in fog |
| Emberscope | Wildfire IR | 5 | 18 | Hotspot detect at 4km |
| Meadowlark | Bioacoustic + optical | 6 | 28 | Species ID at 92% recall |
| Northstar | Star tracker | 8 | 40 | 0.7″ drift, −20°C rated |
We publish our field-day count because that’s the number that separates a lab from a slide deck. Names changed; numbers real.
Bench utilization by hour — heaviest 22:00–02:00.
The bench
Instruments, not renders.
0.7″tracker drift
42dBtypical SNR
−20°rated to (C)
14dunattended log
Where the months go
bench
48% bench, 24% field, 28% integration & docs.
Field-test pass
83%
Programs that clear the field bar on the first deployment.
Schedule risk
Weather is the variable we can’t bench. We build slack for it.
On the bench
The instruments we run.
Cooled sCMOS camera
Monochromator
Vibration isolation table
Thermal chamber −40–85°C
Goniometer stage
Edge GPU rig (Orin)
All instruments are calibration-traceable and logged. We photograph the setup for every result we report.
Bench status
What’s running tonight.
RIGPROGRAMPHASESTATE
OPT-1Frostwatch photometerFieldLogging
OPT-2Northstar trackerBenchNominal
THM-1Quayline LiDARThermalSoak −20°
DSP-3Emberscope IRIntegrationAwaiting part
Two program slots, four rigs. When a slot frees up we post it here first — not on a sales call.
Capability envelope
What the bench can measure.
| Domain | Range | Resolution | Env rating |
|---|---|---|---|
| Photometry | 0.1–10&sup5; lux | 14-bit | −20°C |
| Spectral | 350–1100 nm | 2 nm | IP65 |
| Ranging (LiDAR) | 0.3–120 m | ±2 cm | Fog-rated |
| Thermal IR | −20–550°C | 50 mK | 4 km detect |
| Pointing | full sky | 0.7″/min | −20°C |
Fit
We’re the right lab if…
We’ll refer you out if…
- You need a 21-day sprint — that’s Moonfire, not us.
- It’s a pure web/app build with no sensing.
- You need staff augmentation, not a scoped program.
- There’s no way to run a field test. We won’t sign off on bench-only.
From program leads
What partners say.
“They deployed on our orchard for three weeks in February. The frost call gave us 40 minutes — that’s the difference between a crop and a loss.”
— Grower co-op, Central Valley“Every other vendor gave us a demo in daylight. Deep Star ran it at 2am in fog, which is the only condition that mattered.”
— Port operations“The calibration certificate and runbook meant we could rerun it ourselves. No black box.”
— Instrumentation lead“Two programs at a time is real. We were one of the two, and it showed.”
— Research directorWhat you get back
Data you can defend.
Every program ships raw captures and reduced results side by side, with the reduction scripts and the calibration chain. If a reviewer asks how a number was produced, you can show them — no black box, no “trust us.”
Deliverable mix
Reproducibility
100%
Results shipped with a script that regenerates them from raw.
Data volume
Terabytes logged per field deployment, indexed and searchable.
Raw + reduced
Calibration chain included
Source in your repo
Runbook + certs
Engagement shape
From first call to handover.
Intake call and site questionnaire. We decide if it’s a fit before you pay.
Site survey, measurements, and a scope with pass/fail numbers.
Bench, integration, and firmware — milestones you sign off.
Two-plus weeks of continuous logging on your site.
Instrument, source, certs, runbook. Retainer optional.
Bench & contact windows
| Night bench | 21:00–04:00 |
| Calls | Mon–Thu 13:00–18:00 |
| Field on-call | 24/7 during deployment |
| Friday | Reduction & reports |
Engagement tiers
Scoped work, not hourly drift.
Survey$14k2–3 weeks
- Site visit & measurements
- Feasibility memo
- Scope with pass/fail bar
- No prototype
Programfrom $95k3–9 months
- Bench → integration → field
- Instrument + source in your repo
- Calibration certs & runbook
- Real multi-week field test
Retainer$6k/moafter handover
- On-call during deployments
- Quarterly recalibration
- Priority next program slot
- Firmware maintenance
Deep Star Labs is a kept Clawsum lab. We hold two program slots. When they’re full, we say so.
Before you engage
Straight answers.
How are you different from Moonfire Labs?
Moonfire runs 21-day burns to a rough prototype. Deep Star takes multi-month programs where the deliverable is a validated instrument with field data behind it. Different clock, different bar.
Do I own the IP and the code?
Yes. Source lands in your repository and the instrument is yours. We keep no exclusive rights unless you specifically ask us to maintain something on retainer.
Why do you insist on a field test?
Because sensing that works on a bench and fails in the cold is worthless. We won’t sign off on bench-only results — the field test is where we earn the fee.
Why the night hours?
Optical and low-light work needs a dark sky and a quiet noise floor. Our 21:00–04:00 window is when the measurements are honest. Field on-call is 24/7 during a deployment.
What does the chat here do?
It starts a real engagement conversation — scope, timeline, and whether we have a slot. It’s for the lab’s work, not a form to nowhere.