muthugsubramanian/DocWain-14B-v2
DocWain-14B-v2 is a 14-billion parameter unified decoder-only transformer model developed by DHS IT Solutions, designed for end-to-end enterprise document intelligence workflows. It excels at extraction from diverse document types, cross-document synthesis, and conversational Q&A grounded in RAG retrieval, offering hallucination-resistant responses with uncertainty flagging. The model features a 40960-token context length and is optimized for domain-aware reasoning across HR, legal, finance, and medical sectors.
Loading preview...
DocWain-14B-v2: Unified Document Intelligence Model
DocWain-14B-v2, developed by DHS IT Solutions, is a 14-billion parameter model designed for comprehensive enterprise document intelligence. It integrates document extraction, intelligence-brief generation, multi-document synthesis, and conversational Q&A into a single checkpoint, eliminating the need for separate sub-models or adapters. The model is built on a unified decoder-only transformer architecture with a 40960-token context length.
Key Capabilities
- End-to-End Document Workflow: Handles extraction from various document types (PDF, DOCX, Excel, CSV, images, scanned) and generates intelligence briefs.
- Domain-Aware Reasoning: Excels in enterprise domains such as HR, legal, finance, medical, content, operations, compliance, and security.
- Cross-Document Intelligence: Supports comparison, aggregation, contradiction detection, and ranking across multiple documents.
- Grounded Content Generation: Produces content with named citations, ensuring hallucination resistance and flagging uncertainty.
- Conversational AI: Provides Q&A grounded in RAG retrieval and offers contextual follow-up suggestions.
Intended Use Cases
- Document Q&A grounded in a retrieval index.
- Generating per-document intelligence briefs (headline + key points).
- Cross-document synthesis, comparison, and ranking.
- Providing conversational follow-up suggestions within the DocWain runtime.
Limitations
While robust, DocWain-14B-v2 is not intended for standalone open-domain chat without retrieval grounding or for generating legally binding documents. Its performance relies on the DocWain runtime's RAG layer for grounded enterprise context, and it is recommended for human-in-the-loop assistance rather than unattended high-stakes document review.