An offline data analysis system that extracts, organizes, and indexes information from diverse file formats — including audio — and makes mission-critical content fully searchable. No network access required.
Analysts can inspect originals, extract metadata, identify faces, map geotags, and share findings — without leaving the system.
Documents, archives, images, audio, and email.
OCR, metadata, geolocation, faces, and entities.
Elasticsearch results with direct access to originals.
Interactive geolocation and face match review.
Notes, links, sharing, and collaboration.
Download case output as DOCX.
Modern data science models and LLMs applied according to customer infrastructure, capabilities, and operational needs.
Text extraction and recognition in low-quality images and complex documents; multilingual processing.
Automated extraction, validation, indexing, and unified metadata views across supported file types.
Detection with MTCNN and recognition with InceptionResnetV1 using vector-space comparison.
Whisper v3 with CTranslate2 converts audio to searchable text with language detection.
Metadata search and map-based analytics for GPS and enriched location information.
Originals, notes, sharing, permissions, audit logs, and DOCX export support investigation work.
Multilingual OCR across Latin and Cyrillic scripts — no cloud, arbitrary backgrounds, noise, orientation, and layouts — at 100GB daily input and 30TB historical data, fully automated.
| Configuration | Total time | Time / image |
|---|---|---|
| Tesseract OCR · Intel i7 laptop | 814 sec | 12.7 sec |
| LLM OCR · Intel i5 CPU | 2,895 sec | 45.2 sec |
| LLM OCR · i5 + RTX 4070 Ti SUPER | 394 sec | 6.15 sec |
Two specialized modules extend Snowdrop into visual identity matching and searchable speech.
Detect faces in images and compare embeddings in vector space — search person or document databases, view close-ups and originals, and add matches to cases.
Transforms audio into searchable text with automatic language detection — measured, not promised.
Confirm offline deployment constraints, file mix, and processing targets — and we'll prepare a PF30 demonstration around them.
Request a demo