AI FlowOps
Read-only guided demo of a completed AI-processed workflow.

System design

How AI FlowOps turns CORTEX Presolicitation Review into a governed business decision

The public Case Room is a read-only guided demo of a completed AI-processed workflow. It shows the same operating pattern the backend supports: opportunity screening, intake, extraction, specialist routing, human review, synthesis, decision logging, and KPI capture.

Opportunity Discovery / AI ScreeningMonitor feeds, download notices, filter likely contract-fit RFPs
Document InputsRFPs, contracts, notes
Ingestion LayerParse and normalize
Package ProfilerScore size, page count, amendments, Q&A, and prompt budget
Simple Direct AI PathSmall packages go straight to grounded extraction and routing
Large Normalized-Packet PathDense packages are inventoried, chunked, reconciled, and summarized first
AI Decision EngineExtract, classify, recommend
Workflow EngineRules, routes, packets
Specialist HITL LayerLegal, Security, Finance, Implementation
AI Synthesis LayerCollate reviews and prepare BD/Ops packet
BD/Ops HITL LayerOwn final bid/no-bid decision
Database / Audit LogCases, events, traces
Dashboard LayerKPIs and proof views

Dual-path opportunity processing

Simple path

Small package, direct AI review Clean opportunity packets stay fast: extracted documents become one intake package, then AI routes Legal, Security, Finance, and Implementation packets.

Large path

Dense package, normalized master packet Large government packages are inventoried, classified, chunk-reviewed, reconciled, and converted into an AI-normalized packet before routing.

Traceability

Every normalized fact keeps its source Findings retain source filename, page range, quote, normalized fact, confidence, and chunk ID when chunking is used.

Presolicitation frame

Pursue/prepare, not bid/no-bid Presolicitations use a more permissive capture-readiness prompt, while final solicitations keep bid-stage decision language.

Technical stack

  • FrontendServer-rendered Jinja templates and repo-owned CSS
  • BackendFastAPI application routes and workflow APIs
  • DatabaseSQLite persistence for cases, traces, approvals, and KPIs
  • AI workflowAI processes the source documents into screening, analysis, packets, and synthesis; public pages display the completed run read-only

Deployment flow

Local Development GitHub GitHub Actions Private VM Demo

What this proves

Product design

End-to-end operations workflow One business question moves through AI, specialists, BD/Ops, and audit records.

AI system design

Grounded decisions Every major claim links back to a source phrase and an extracted fact.

Workflow automation

Department-specific packets The AI generates different packets for Legal, Security, Finance, and Implementation.

Governance

Human-owned decision AI recommends; specialist reviewers and BD/Ops own the final decision.