The Lab
Build it. Examine the result. Decide what is useful.
The Lab is where I explore practical questions through working systems, research workflows and prototypes. I test what they produce, examine where they fall short and use that learning to inform how I approach operating problems.
Some builds support my own work. Others investigate a method or technical approach. The featured builds below can be looked inside.
Featured builds · Account Intelligence · AI-assisted research workflow
What does this signal actually justify?
What it is for
Bring useful account evidence together before deciding where to put commercial attention.
What it does
The research combines public sources with available account context and returns a structured assessment: what was observed, why it might matter and what remains uncertain.
Why it is useful
It gives me something to examine before deciding whether to take a next step. A visible role, company change or apparent requirement can justify investigation without establishing that someone is ready to buy.
What you can inspect
The selected example contains conflicting evidence about whether a requirement is still open. Inspect the output, the evidence, the recommended check and the boundary around the decision.
Inspect the exampleThe useful distinction
Research can recommend a next check. The commercial decision still belongs to a person: ACT, HOLD or no meaningful signal. The selected output does not record that final decision.
What this example leaves to test
A useful next test is whether the assessment saves review effort while retaining the uncertainty that matters. The example does not establish a productivity gain or an automated CRM action.
Trigebra · HubSpot diagnostic build
Can the finding be traced back to the system evidence?
What it is for
Examine a HubSpot environment and produce diagnostic findings a practitioner can check.
What it does
Trigebra reads portal and configuration data through the HubSpot API. Deterministic checks collect evidence, and an LLM synthesises findings within those constraints. Confidence treatment and QA sit alongside the result.
Why it is useful
A finding is more useful when I can see the observation behind it, understand its limits and decide what it means for the operation. If the available information does not support an assessment, an unassessable result is valid.
The design choice
The model synthesises evidence collected by the diagnostic. It is not used to invent the underlying portal state. That separation gives the practitioner a basis for checking the finding.
What you can inspect
- 01HubSpot API
- 02Configuration evidence
- 03Deterministic checks
- 04Constrained synthesis
- 05Confidence and QA
- 06Finding
What remains a test question
Does the finding identify something useful, and is its confidence justified by the available evidence? Those questions require evaluation of actual outputs, not just a plausible architecture.
From the Lab to an operating problem
I use the Lab to investigate an approach before deciding where it belongs. For client work, the requirements, access, evaluation and operating ownership still need to fit the actual situation.
Have a question the Lab could help answer?
Bring the system, workflow or decision you are trying to improve. We can establish whether an approach is worth building and how it would be evaluated.