How do we send documents to super-gpu?

Upload with SCP behind a VPN to a URL super-gpu provides; give super-gpu a URL, message queue or API to pull the images from your server; ship a NAS or DAS to the super-gpu location; or call the super-gpu AI services directly from an API.

How are results returned?

By SCP behind a VPN directly to your server, through your own API, as SCP download details behind a VPN, or on the NAS or DAS you provided, returned with the results on it.

Can we bring our own model?

Yes. You can bring your own model to run on super-gpu hardware, and super-gpu AI advisors can help select the best model, pipeline or hardware for the process.

Which document formats are supported?

Any operating system, encoding and document format is supported.

Input type — Documents

AI models for documents

Twenty-five services that read a document and return structured output — from a scanned invoice to a contract archive. One file over an API, or a whole repository in a batch, on the same dedicated cluster.

The right model for the job

Our AI advisors may assist you to select the best model or pipeline for the process or offer the most efficient hardware to perform the process.

Of course, you can bring your own model to run on our hardware.

In case you like to use your on-premise servers cluster or a certain cloud we may provision the needed services and build the pipeline.

One model chosen from the catalogue, matched to your workload and hardware

AI services for documents

Our service catalogue include vast selection of services.

Part of the services based on stable AI models and part of the services based on our native code.

Each service can use several models and methods. After a short discussion we can decide on the right pipeline for the process.

Find a service

Understanding and answering

Private chatbot / enterprise assistant: An assistant that answers from your own documents, inside your own deployment.
Document question answering: Ask a question of a document set and get the answer with its source passage.
Text summarization: Long documents reduced to a brief, at the length you ask for.
Document visual question answering: Ask a question of a page image and get the answer with the region it came from.

Classification / analysis

Text classification / tagging: Documents and passages sorted into your own label set.
Document classification: Each page or file sorted into your own document types.
Sentiment / intent analysis: Tone and intent scored across correspondence, feedback and tickets.
Customer-support ticket routing: Incoming tickets classified and routed to the right queue.
Content moderation - text: Text checked against a policy you define, with the reason recorded.

Extraction

Named entity recognition: People, places, organisations and dates located in running text.
Fact / relation extraction: The relationships between entities pulled out as structured statements.
Structured data extraction from text: Free text turned into the fields your systems expect.
PDF/document to structured JSON: A whole document parsed into JSON, layout and hierarchy preserved.
Invoice / receipt extraction: Supplier, dates, line items and totals as fields, not free text.
Form extraction: Key-value pairs pulled from structured and semi-structured forms.
Table extraction: Printed tables recovered as rows and columns, ready for a spreadsheet.
Contract analysis / clause extraction: Clauses, parties, terms and obligations identified across an agreement.

Search and compare

Semantic search: Retrieval by meaning rather than keyword, across your whole corpus.
Search reranking: Candidate results reordered so the most relevant sit at the top.
Duplicate / near-duplicate text detection: Repeated and lightly edited copies found across a large collection.

Language / rewriting

Multilingual translation: Documents translated between the languages your material uses.
Grammar / style rewriting: Text corrected and rewritten to a house style you set.

Privacy / hidden data

PII detection and redaction: Personal and sensitive data found and masked before anything is shared.
Document redaction: Names, numbers and chosen regions masked before a document is released.
Image steganography - hide text in photo: A message carried inside a photograph, invisible to the eye.

How to provide us documents?

We support any OS / Encoding and format for documents.

You can upload your documents with SCP (behind VPN) to a URL we provide.

You can provide us URL, MQ or any API and we will pull the documents directly from your server.

You can send a NAS or a DAS to our location.

You can use our AI services with an API.

How to get the results?

We can provide the results with SCP (behind VPN) directly to your server, to use your API for results, send you SCP (behind VPN) details to download the results or send you back the NAS/DAS you provided with the results.

Other input types

Pictures Faces, objects, counting, captioning, quality, hidden data. AI models for Pictures →
Video files Tracking, transcription, subtitles, summarization. AI models for Video files →
Audio files Transcription, diarization, separation, summarization. AI models for Audio files →
Live feed Counting, tracking, live transcription, PPE and queues. AI models for Live feed →
Text files Summarisation, extraction, translation, PII detection. AI models for Text files →