Private AI, air-gapped VPC. AI Safety & Data Security guaranteed. One Hub for business and technical users โ conversational, in plain English.
Year 1 wasn't a rollout. It was foundation work โ the underlying capabilities every Year 2 use case depends on. Eight pillars, all delivered.
xAQUA deployed on the Authority's own AWS. Monitoring across GPU, EC2, EKS, RDS, Lambda. Auto-scaling, Splunk logging per ISO requirements, 24ร7 availability.
A private LLM running entirely inside the agency's own VPC. No external APIs, no telemetry, no inference touching the public internet. Open-source LLM cuts cost without compromising the boundary.
Every LLM call carries customer ID, user ID, use-case context, and sub-context โ per-use-case routing, audit trails, rate limits, policy enforcement. Private AI safely usable by many applications at once.
RBAC, audit logging on every interaction, PII masking applied uniformly, ISO-aligned controls. The architecture has no path to external infrastructure โ that's the design, not a setting.
Core benefits database cataloged with technical, business, and semantic metadata. 450+ relationships auto-discovered. The shared layer every agent inherits.
Multi-dimensional profiling โ completeness, uniqueness, consistency, validity. Custom rules engine with automated alerting. The framework Year 2 builds the formal practice on top of.
Architecture for Archer, the agency's standalone RAG-powered AI assistant. User auth, RBAC, selective retrieval, source transparency, chat audit logging. The pattern every Year 2 assistant inherits.
Secure data harvesting with RBAC, PII masking, audit logging, and multi-format export. NLQ-to-SQL over the core benefits database. The plumbing the 13โ15 analyst team is being phased onto in Year 2.
Purpose-built assistants for the leaders of a $400B+ fund. Same governance, same audit trail, every output cited.
22 use cases across board meeting preparation, strategic plan monitoring, modernization program oversight, audit & risk intelligence, legislative tracking, actuarial briefings, stakeholder communications, workforce intelligence, committee synthesis, and real-time decision support during board meetings and media inquiries. 35โ55 hours/week of CEO-level time savings.
Investment portfolio oversight, IPS compliance, allocation analytics, asset-class drill-downs, manager performance, risk posture. Conversational access to the data and decisions that shape a $400B+ portfolio.
For senior leaders across the investment branch: collaborative model analytics, peer benchmarking, portfolio risk reports, semi-annual CIO activity intelligence. Same-second answers on the questions that drive trustee oversight.
Funding plan trajectory, scenario analysis, return assumption sensitivity, funding risk reports, modernization-readiness scoring. The CFO/Actuary's plain-English line into the fund's long-horizon math.
Not a chatbot. Not an OCR vendor. The unified document intelligence platform the Frankenstack of point tools has been pretending to be. One ingestion. Seven destinations.
Grounded Q&A with page-level citations on every claim. Hybrid BM25 + vector retrieval. Per-domain agents (Pension, HR, Legal, Finance, IT). Effective-date awareness for amended documents. Refuses to fabricate if retrieval is below confidence.
Generate SOPs, BPAs, RACI matrices, test cases, requirements specs, briefing memos, audit responses, and board papers โ from your sources, in your templates. Four-stage wizard with an approval state machine and immutable audit trail.
Turn PDFs, scans, emails, forms, contracts, transcripts, HTML, JSON, XML, and spreadsheets into clean structured data with confidence scores you can route on. Custom schema designer. Low-confidence rows auto-route to human review. Inline corrections save back as ground truth.
Automatically identify and mask sensitive data across your corpus. PII, PHI, financial identifiers, and custom sensitivity patterns you define. Multiple modes โ redaction, pseudonymization, tokenization, format-preserving encryption. Role-based selective reveal. Every mask logged.
The governed home for every output, template, and schema. Versioning, supersession chains, pinning, cross-collection search. Permissions inherited from source-system ACLs. Audit-ready by default.
Coverage check, risk scan, redline draft in ~90 seconds. Bootstraps a playbook from your historical contracts. Counterparty-specific learning suppresses recurring false positives. Tracked-changes DOCX export โ send straight to counterparty.
Living map between regulations (HIPAA, SOX, GDPR, FedRAMP, ISO 27001, PCI-DSS, ERISA, custom obligations) and your policies, procedures, controls, and contracts. Audit-evidence package generation in seconds.
A document indexed for Chat is instantly a source for Create. A correction in Extract becomes ground truth for the next Chat answer. A masking policy applied once protects every downstream function. Shared substrate, not stitched APIs.
DocIQ runs on documents wherever they live โ S3, Azure Blob, GCS, SharePoint, OneDrive, Google Drive, Box, Confluence โ at the same time, with one search, one permissions model, one audit log. Your bytes never leave the source.
Beyond the executive portfolio and DocIQ โ four more initiatives that change how the agency operates day to day.
The 13โ15 analyst SQL team being phased onto AI-assisted query generation. Three-layer QA โ automated harness validation, tiered human review, continuous feedback. Sensitive queries always require human sign-off. Analyst expertise redirected to complex analytical work.
Rule library across six quality dimensions, monitoring dashboards, steward workflows, governance with domain ownership and remediation SLAs. Modernization-readiness scoring by domain.
Natural-language querying (ConverseDataIQ), automated narrative reporting (Narratix), and an Analytics Data Lake. Department-by-department rollout across actuarial, member services, employer reporting, compliance. Predictive analytics into actuarial scenario modeling.
Live operational intelligence: call-center recovery, member adoption, defect burn-down, M&O budget. Automated AI enrichment of post-go-live incident records โ categorization, clustering, root cause analysis, Known Error Database, release prioritization.
In Year 1 we established the data and AI foundation โ private AI inside the agency's air-gapped VPC, the LLM Gateway, the semantic catalog over 700+ tables, the data quality framework, secure data harvesting, and the standalone RAG architecture for Archer. In Year 2 we scaled that foundation into the business โ executive AI for the CEO, CIO, and investment and actuarial leaders; the full DocIQ document intelligence suite; analyst-team augmentation; the agency's first formal DQ practice; and live operational intelligence across the $500M+ modernization program.
If you're a public agency, regulated enterprise, or modernization leader looking at AI that respects your data residency and audit posture โ we should talk. The platform is designed for the bar this story sets.