Knowledge ingestion
Enterprise documents and knowledge sources are ingested, processed and prepared for semantic retrieval.
ENTERPRISE KNOWLEDGE, GROUNDED IN EVIDENCE
EvidNexa securely connects enterprise content, searches across it with AI and grounds answers in the source material used to produce them.
KNOWLEDGE ASSISTANT
Responses are generated from authorised enterprise knowledge and accompanied by supporting evidence.
THE PRODUCT
EvidNexa demonstrates how retrieval-augmented generation can be integrated into an enterprise application without separating AI capability from identity, authorisation and data isolation.
Enterprise documents and knowledge sources are ingested, processed and prepared for semantic retrieval.
Retrieval provides relevant context to the language model and exposes supporting citations alongside generated answers.
Knowledge access is scoped by tenant so users cannot retrieve another organisation's protected content.
OpenID Connect authentication and role-based authorisation protect application and administrative capabilities.
Administrative capabilities provide controlled access to operational and tenant-level functions.
Application logging, request tracing context and cloud monitoring provide visibility into runtime behaviour.
AWS ARCHITECTURE
EvidNexa combines a React interface, Python API, PostgreSQL vector retrieval and AWS managed services in a containerised architecture designed around security and tenant isolation.
SECURITY BY DESIGN
The application is designed so retrieval, administration and application access operate within explicit identity and tenant boundaries rather than treating the AI layer as a separate trust zone.
Enterprise authentication through Amazon Cognito and standards-based OpenID Connect.
Viewer, editor and administrative capabilities are separated through application roles.
Retrieval and document access are constrained by tenant boundaries.
Persistent application data is kept behind private network boundaries.
AWS managed encryption protects persistent application and document data.
Security-sensitive application activity can be recorded with actor and request context.
ENGINEERING CASE STUDY
EvidNexa combines enterprise RAG with the engineering required to operate it securely: identity, tenant isolation, private networking, deployment automation, failure handling and cost control.
Tenant-scoped vector retrieval, reranking and grounded generation return answers with supporting citations.
Cognito OIDC, RBAC, PostgreSQL RLS, Pod Identity, Secrets Manager, KMS and temporary STS credentials protect the application and its data.
EKS, private RDS, ECR, Kubernetes and Terraform provide a reproducible cloud platform without exposing the database publicly.
GitHub Actions publishes immutable container builds even while EKS is offline and deploys automatically when the runtime is active.
Bedrock throttling and quota failures are classified and handled cleanly rather than exposing raw SDK errors or corrupting ingestion state.
The public portfolio remains online while EKS, RDS runtime resources, NAT and the application load balancer can be destroyed between demonstrations.
VERIFIED ENGINEERING RESULTS
LIVE APPLICATION
The production-style AWS environment is operated on demand rather than kept permanently running. This keeps the portfolio cost-efficient while preserving the complete application and infrastructure implementation.
A live environment can be made available for technical walkthroughs, architecture discussions and demonstrations.
Contact via LinkedIn