Ask your data in plain language. Get answers you can check.
SACPKG keeps a very large knowledge graph in one small file and answers questions straight from it. Algorithms compute the answer from your records, and every name and number is checked against the data before you see it.
Four steps, from your data to a checked answer
Your data, one file
Your records become a knowledge graph, compressed into one file on your own machine, with nothing lost.
Ask in plain language
AI reads the question and works out which records and links it needs.
Computed, not guessed
Algorithms find the records, follow the links, count and compare.
Checked before you see it
Every name and number in the answer is checked against the data, and each record can be traced to its source.
Every trial on ClinicalTrials.gov, on a laptop
We loaded the entire public registry, 608,000 trials, into SACPKG.
A real question
only drugs, not other interventions, for Schizophrenia in the final phase
SACPKG read this as: Schizophrenia trials, keep phase 4, their interventions of type drug, list the names. It returned the complete list of 296 drugs in 13 seconds on a laptop, each one traced to a stored trial record.
Accuracy
On our test sets of questions about specific records, from banking, film and clinical data, SACPKG answered 92 to 100 percent correctly. When it missed, it never made up a record: the misses were misread questions, wrong counts or the wrong format.
We are extending it to open, analysis-style questions across whole datasets.
Runs where your data lives
Smaller teams can run SACPKG on one laptop-class machine with small local models. Larger organisations run it on their own servers, with bigger models for harder questions. A hosted option is planned for teams whose data can be hosted.
Private data never has to be sent to an outside AI service.
- Pharma and researchTrial and research data on drugs, outcomes, sponsors and safety.
- Banks and insurersCustomer, account and transaction records that must stay in-house.
- GovernmentRegistries and public records, queried on its own servers.
See SACPKG on data like yours
We start with a short paid pilot on your own data, measured against how you work today.