Evaluation of the claim
The product run against a harness built for its task, with error rate and failure modes reported.
Model dependency and margin
Which providers the company depends on, what switching costs, and what inference does to gross margin.
Data rights and defensibility
Where the training and operating data comes from, who owns it, and whether it compounds.
Architecture and team review
Whether the system is built to be verified, governed and scaled, and whether the team can do it.
Industries/Capital and risk
Gallahand forVenture Capital
Separate the AI companies that compound from the ones that rent a model. Technical diligence, physical AI sourcing, and a shared stack that portfolio companies inherit on day one.
Measured in: evaluated error rate · inference margin
Diligence · EvaluationTask suite · 2,400 cases
Pass 91.4%Threshold 95%
RUN · 2,400 cases complete
The constraint
In AI, the moat is in the verification harness, not the demo.
Depth
Every deck says AI. Few funds can test whether the system holds up under evaluation.
Dependency
Margins and roadmaps sit on another company's pricing, policy and uptime.
Physical AI
The next wave runs in atoms and electrons. The sourcing networks are operators, not demo days.
Use cases
What we build for venture capital.
Each use case runs on the same stack, so the second one costs a fraction of the first.
VENTURE CAPITAL / 02
Physical AI sourcing and thesis
Deal flow from the field, not the conference circuit.
Thesis work
Sector theses across bits, atoms and electrons, written by people who operate in them.
Field deal flow
Founders and problems surfaced from the operators and assets we already work with.
Proving grounds
Portfolio companies tested on real fleets, sites and energy assets, under agreement.
Co-build
Gallahand builds alongside founders where the physical integration is the hard part.
VENTURE CAPITAL / 03
Portfolio acceleration
Each company starts with the stack, the envelope and the customers.
Shared intelligence stack
GRAIL routing, evaluation and policy envelope, inherited rather than rebuilt.
Evaluation as a service
A verification harness that ships before the agent does, kept current as models change.
Route to physical operators
Introductions and pilots with the operators who buy from this category.
Portfolio telemetry
Operating data that informs follow-on decisions, viewed in SIEGE.
Integration
Diligence work is performed under NDA and delivered as a written memo with the evaluation data attached. Fund relationships are never referenced publicly.
Platforms in play
One stack underneath.
GRAILLayer 01 · Intelligence
The intelligence core. Model routing, agents, evaluation and the policy envelope, on your compute or ours.
CORBENICLayer 02 · BitsThe operating picture. Telemetry, state and memory joined into one live model of every asset.
AETOSLayer 03 · Atoms · AerialAerial intelligence. Dock-based survey, inspection and overwatch, interpreted the same hour.
SIEGECapital layer · By invitationThe portfolio, seen from one seat. Asset telemetry rolled up into evidence a board can act on.
How an engagement runs
One number, moved and proven.
01 · SURVEY
Find the constraint
Two to four weeks on the asset. The number that costs most, the data that already exists, and what has to work first.
02 · INSTRUMENT
Connect what runs
Existing systems connected and gaps closed with vendor-neutral hardware. A baseline both sides sign.
03 · PROVE
Move one number
A bounded deployment against the baseline, reported in physical units, with a human gate on anything irreversible.
04 · OPERATE
Run it, or hand it over
We operate the system with your team, or transfer it with the code, the models and the playbook.
We do not publish client names or case studies. References are shared in a briefing, under agreement.
From the thinkingThe Aircraft Is Not the ProductIn an air network the margin sits in a layer no regulator certifies.Read ↗Briefings
Send the constraint, not a brief.
Most of our work arrives by introduction. If you run or back capital in this space, tell us the number that is costing you most.
- 01Your stage, cheque size and thesis.
- 02A company or claim you want tested.
- 03The sectors in atoms and electrons you want to own.
- 04What your portfolio companies most need from a shared stack.