How the Tech Works

We use large language models (Amazon Bedrock) combined with a structured knowledge base to analyze veteran records against VA rating criteria.

The system doesn't guess — it retrieves relevant regulations from our database of 38 CFR, BVA decisions, CAVC rulings, and the M21-1 adjudication manual, then cross-references your specific records.

Every finding cites the specific regulation or precedent. If we can't cite it, we don't show it.

Our system is built on every public VA data source available:

• 38 CFR Part 4 — Schedule for Rating Disabilities (all diagnostic codes)
• Board of Veterans Appeals (BVA) decisions
• Court of Appeals for Veterans Claims (CAVC) rulings
• M21-1 Adjudication Manual (VA's internal procedures)
• PACT Act guidance and presumptive conditions
• OGC Opinions and Federal Register updates
• Secondary conditions mapped with PubMed medical research

This isn't a generic chatbot. It's a retrieval-augmented generation (RAG) system with over 6,000 verified knowledge chunks, vector-embedded for semantic search.

Upload a DD-214, medical records, C-File, or decision letters. The system:

• Extracts text using AWS Textract OCR (handles phone photos of documents)
• Identifies document type and structures the data
• DD-214: extracts MOS, decorations, service dates, combat indicators
• Medical records: identifies conditions, medications, diagnoses
• C-File: processes the entire file and finds evidence for conditions
• Decision letters: parses granted/denied conditions with ratings

The deep analysis cross-references all your uploaded records against our full knowledge base to find:

• Conditions supported by your evidence but not yet claimed
• Secondary connections (one condition causing/aggravating another)
• Evidence gaps — what you'd need to strengthen a claim
• Presumptive conditions based on service era, location, or MOS
• Rating criteria and what evidence maps to which level

Results are deterministic — the same records produce the same analysis. No hallucination, no random outputs.

Your data is sensitive. We treat it that way:

• All data encrypted at rest (AES-256) and in transit (TLS 1.2+)
• AWS GovCloud-grade infrastructure (us-east-1)
• Mandatory MFA on all accounts
• Session timeout after 1 hour of inactivity
• ISO 27001-aligned security policies
• SOC 2-ready controls (access control, audit logging, incident response)
• Your documents are never used to train AI models
• Delete your account at any time — all data permanently removed