Conduit
CDTIngest and normalisation
Heterogeneous formats, encodings and metadata schemas resolved into one typed, addressable record. Partial submissions, inconsistent naming and duplicates across sources are detected at the point of entry.
Four software modules that take images, documents and their metadata, resolve them into records a downstream system can rely on, and score the result so the same input produces the same judgement tomorrow. Delivered as fixed-scope software against written acceptance criteria.
At volume, incoming files contradict their own metadata. Fields are missing. The same study appears twice under two identifiers. An image has been resaved, cropped or regenerated somewhere between the source and you.
A person can catch any one of these. No team catches all of them at volume, and a person who checks the same thing twice does not always answer the same way. This is the layer before analysis, not analysis.
Ingest and normalisation
Heterogeneous formats, encodings and metadata schemas resolved into one typed, addressable record. Partial submissions, inconsistent naming and duplicates across sources are detected at the point of entry.
Verification and quality control
Rule-based and model-based checks on incoming data, provenance and manipulation detection on images, and a review queue that routes only what the automated checks cannot settle.
Matching and retrieval
Structured and semantic retrieval across normalised records, where every ranking carries the reasons for it.
Evaluation harness
A fixed test corpus and a scoring method, so changing a model, a prompt or a threshold produces a number comparable to the last number.
Baseline exists because the honest answer to how you know the model still works cannot be a demonstration: any vendor can show a good result once. It has to be a measurement, taken the same way every time, under change control, available to whoever asks.
The pipeline is domain-neutral; the configuration is not. The modules encode no clinical knowledge. What makes them useful is the configuration: the checks, thresholds, schemas and acceptance tests that encode what your specialists already know. That configuration belongs to you.
The first configured domain is clinical research and imaging. The documentation describes what each module accepts and emits, and the company page describes where the pipeline already runs.
A format the modules have not seen is a configuration change, not a rebuild. Conduit resolves a new source into the same typed record by a written schema and a set of checks; Assay runs the rules and models that record is scored against; Baseline holds the corpus that says whether the new configuration performs the way the last one did. What that costs is scoped and quoted against acceptance criteria like any other module, and if it cannot be scoped that way you will be told so.
Acceptance criteria are agreed in the statement of work before work begins, and Quantscope carries the delivery risk. Not time and materials. Not people placed inside your team.
Send the data type, the volume, and what currently has to be checked by hand to info@quantscope.ai. If a module cannot be scoped against acceptance criteria, you will be told so rather than sold an engagement built around it. Response within two business days.