Engine Architecture — 5 Layers
L-0
RaggrIngest
PDF · Markdown · DOCX · CSV · TXT · JSON · Smart PDF routing (local extraction or an OCR service)
6 FORMATS
↓
L-0.5
RaggrKairo
Intent classification: RETRIEVE · GENERATE · COMPARE · SUMMARISE · Query enrichment before retrieval
INTENT ROUTING
↓
L-1
RaggrCore
Keyword + semantic search · Rank fusion · Reranking
HYBRID SEARCH
↓
L-2
RaggrGuard + RaggrPII
PII detection · Citation enforcement · Confidence threshold gate · Prompt versioning · Audit log
CITATIONS
↓
L-3
RaggrFormatter
Structured, domain-aware, intent-specific answers
STRUCTURED OUTPUT
Guardrail Capabilities
PII Detection
Citation Enforcement
Confidence Threshold Gate
Audit Log
Prompt Versioning
Intent Classification Log
Product Portfolio — One Engine
RAGGR™ Enterprise
GCC knowledge intelligence. Multi-tenant. Whitelabel.
BōdhanAI
DEMO ↗
Class 11–12 EdTech. NCERT + Samacheer. Tamil + English. "Understand. Not memorise."
Multi-Tenant Architecture
Platform
└→ Tenant: Example
└→ Domain: example_domain_a
└→ Pool: pool_a
└→ Pool: pool_b
└→ Domain: example_domain_b
└→ Pool: pool_c
Knowledge pools are organised per tenant · Per-tenant cost attribution
Platform Modules
Core engine
RaggrCore✅ Implemented
Hybrid retrieval with reranking and cited answers
RaggrIngest✅ Implemented
Upload, batch intake, chunk preview and maker-checker publication
RaggrVault✅ Implemented
Original-file access and comparison
Safety and governance
RaggrGuardIn development
Role-based access, tenant scoping and citation enforcement
RaggrPIIIn development
PII detection and masking before model calls
RaggrKairoIn development
Query routing by intent and sensitivity
Quality and operations
RaggrEval✅ Implemented
Benchmark harness and quality checks
RaggrLogIn development
Usage, cost and audit logs
Next
RaggrTOCIn development
Structure-aware retrieval
RaggrGraphIn development
Entity and relationship retrieval
RaggrConductorPlanned
Role-scoped skills and bounded agents
Compliance workflows
Consent evidencePlanned
Records the purpose, notice version and consent behind each use of data.
Notice drafting with reviewer approvalPlanned
Structured notice drafts with source links and reviewer approval.
Grievance assistancePlanned
Case intake, owner assignment and reviewed response drafts.
Retention monitoringPlanned
Tracks retention information, with holds and erasure to follow.
Rights-request handlingPlanned
Access, correction, erasure and nomination cases with verified identity.
Status reflects current development. Planned items are directions, not commitments.
Maker-Checker Ingestion Pipeline
📤
1. INTAKE
File upload · type detection · original preserved · duplicate detection
🔍
2. EXTRACTION
TOC · schema · hierarchy · PII scan · content type classify
✏️
3. MAKER REVIEW
Include/exclude sections · tag domain · set pool · add notes
✅
4. CHECKER APPROVAL
Read-only preview · approve → ingest · reject → back to maker
⚡
5. INGEST
Approved sections → chunk → embed → index · audit log written
Retrieval side — pool-scoped access:
Query arrives → access check → the user's allowed pools are resolved → intent classified →
hybrid search scoped to the pool → rerank → filter to approved documents → guardrails → answer with source citations