What document types does ATC Manthan support?
+
PDFs, Microsoft Word, scanned images (via OCR), PowerPoint, Excel, plain text, EPUB, and HTML. Scanned documents are OCR-processed before indexing, with confidence scores attached to extracted text.
How does ATC Manthan avoid hallucination?
+
Every answer is grounded in a retrieval step over your indexed documents. If the retrieval confidence falls below threshold, the system replies "not found in your documents" rather than inventing an answer. Every answer carries citations to the source page.
Can it run completely offline?
+
Yes. After initial deployment and model loading, ATC Manthan runs fully air-gapped with no outbound network calls. Updates are delivered via signed offline packages.
Which languages are supported?
+
Over 100 languages including English, Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Arabic, French, German, Spanish, and Mandarin. Cross-lingual retrieval is supported (ask in Hindi, retrieve from English documents).
What hardware does ATC Manthan need?
+
A single server with 1×A100 80GB or 1×H100 GPU handles up to 100 concurrent users and indexes up to 100,000 documents. Larger deployments scale horizontally with shared object storage and a vector DB cluster.
Is ATC Manthan DPDP-compliant?
+
Yes. All document storage, indexing, and inference happen on your infrastructure. Audit logs of every query and access event are tamper-evident and exportable for DPDP, HIPAA, or RBI compliance reviews.
How long does deployment take?
+
For pre-configured Quick-Start deployments: 3 to 5 days from hardware ready to live. For custom integrations with existing DMS, SSO, and audit pipelines: 4 to 8 weeks.
