Work

Everything here is labelled for what it is. We have not yet delivered a client project, so there are no case studies; when there are, they will be shown only with the client’s permission.

Demos with measured results

Demo data is synthetic; numbers are measured, not promised.

Synthetic data

FatturaPA reader and SdI receipt explainer

Italian e-invoices (XML and signed .p7m) to one checked spreadsheet, where the implemented checks flag defined problems, with the SdI rejection code where one applies; SdI notices explained in plain words.

Test results and limits

On 3 synthetic sets of 94 invoices: 28 of 28 planted faults detected in each, 0 of 66 clean invoices flagged by mistake (measured 5 October 2026).

Read the project: FatturaPA reader and SdI receipt explainer
Synthetic data

Ask your documents

Questions answered from your own files with the exact sentence and its source, offline, with rules to refuse unsupported questions.

Test results and limits

On the synthetic hold-out set: 36 of 43 correct; 11 of 12 unanswerable questions refused and 1 answered wrongly (measured 5 October 2026).

Read the project: Ask your documents
Synthetic data

doc2data

Invoice and receipt PDFs and scans to checked, structured data.

Test results and limits

0 field errors on 3 clean synthetic sets of 16 documents; 92.9% (131 of 141) and 91.4% (127 of 139) of fields on degraded synthetic scans (measured 2 October 2026).

Read the project: doc2data

Test evidence

Tests, data and limits, one project at a time.

TEST REPORTdoc2data test evidenceEvery doc2data number, by set and field by field, with how the sets were built.
Synthetic data

doc2data

Measurements and limits

0 field errors on 3 clean synthetic sets of 16 documents; 92.9% (131 of 141) and 91.4% (127 of 139) of fields on degraded synthetic scans (measured 2 October 2026).

Read the project: doc2data

Open source and research

Open source

Bayesian-Classifier-Lab

A reproducible Python framework that compares five classifier families with paired repeated cross-validation, a Bayesian correlated t-test and a region of practical equivalence, to show whether a difference between models is large enough to matter.

Open source

PsychoGraph-Net

A research model that combines a graph encoder with a temporal Transformer, with LSTM and CNN baselines and a perturbation-based explanation method. It runs on synthetic patient-like data and is not a clinical tool.

Bring your workflow into focus.

Send up to 3 samples or describe one process. We reply by email with what is feasible and how we would measure it.

Start a project
SNELLO / TOOLS

This explains the tool; it is not a live AI session. No document is uploaded.

Our working method
Full diagram

Read the project

Discuss your project

Send up to 3 samples or describe one process, the tools you use and the result you need.

snello.contact@gmail.com

Go to the Contact page

Navigation