Nexus — Use Case: ENDO AI Three Pipelines

Section 01 -- Overview [M]

Field	Value
Use Case ID	UC2
Client Reference [M]	ENDO AI by Pryzmat Media (endo.pryzmat.media) -- Polish SMB AI automation product
System Name [M]	ENDO AI Three Pipelines
Domain [M]	AI Automation / Lead Management / Document Processing / Executive Briefing
One-Line Description [M]	Three independently switchable AI employee pipelines (Lead Qualification, Document Lifecycle, CEO Briefing) deployed as a single Nexus org
Status [M]	Pipeline defined -- Demo queued
Target Delivery [M]	Q3 2026 (demo-complete)

Summary [M]

The Problem: Polish SMBs spend 20-40% of staff time on repetitive intake and routing tasks: qualifying leads from incoming SMS/calls, sorting and categorising received documents, summarising voice memos for executive review. These tasks are manual, error-prone, and do not produce audit trails or structured data for analysis.

The Nexus Solution: ENDO AI positions these tasks as autonomous AI employees. Each "employee" is a Nexus pipeline definition + ChameleonV2 form schema + deltaPrism entity class. The client operates three pipelines independently from a single Nexus instance, switching between them via the pipeline profile switcher. Marcin qualifies SMS leads and books appointments. Irena OCRs documents and routes them to the correct department. Przemek transcribes voice notes and produces structured CEO briefings with action items.

Org Structure [D]

D-06 Resolved: Single Nexus org for ENDO (Pryzmat Media). All 3 pipelines within one org. Demo seeder creates 1 org + 5 RBAC users (Admin, Operator, Reviewer, CEO, System). Pipeline profile switcher demonstrates multi-pipeline capability directly on the canvas.

Section 02 -- Actors [M]

Role [M]	Description [M]	Pipelines [D]	Access Level [D]
Admin	ENDO system administrator. Configures pipeline thresholds, RBAC, and integration endpoints.	All three	Full config access, user management, pipeline enable/disable
Operator	Day-to-day staff. Uploads documents (Irena), records voice notes (Przemek), views lead list (Marcin).	All three	Upload forms, read-only lead list, voice note submission
Reviewer	Resolves low-confidence document classification in Irena pipeline.	Irena only	Reviewer ChameleonV2 form: category override decision
CEO	Approves and optionally edits briefings in Przemek pipeline.	Przemek only	CEO ChameleonV2 form: approve/revise briefing + edit action items
System (Automated Nodes)	SMS polling, OCR, AI transform, scheduling, forwarding nodes. No user-facing forms.	All three	Pipeline-internal only

Section 03 -- User Stories [M/D]

Pipeline: Marcin (Lead Qualification)

P1 [M]: As the System, I poll the SMS gateway inbox and create a lead entity for each new inbound message so no lead is missed
P1 [M]: As the System, I auto-qualify each lead against threshold rules (budget >= 5000 PLN, need in scope, not blacklisted) and route to appointment scheduling or polite decline
P2 [D]: As an Operator, I view the qualified lead list with scores, appointment links, and reply status so I have full visibility
P2 [D]: As Admin, I override lead qualification (re-qualify a declined lead) when business context requires it

Pipeline: Irena (Document Lifecycle)

P1 [M]: As an Operator, I upload a document via ChameleonV2 form so Irena processes it automatically
P1 [M]: As the System, I OCR the document, classify its type (Invoice/Contract/Certificate/Correspondence), and route to the correct department mailbox
P1 [M]: As a Reviewer, when OCR confidence is low, I see the document thumbnail, category suggestion, and confidence score, and I confirm or override the classification
P2 [D]: As a Compliance user, I view the full document processing audit trail per entity in deltaPrism

Pipeline: Przemek (CEO Briefing)

P1 [M]: As an Operator, I upload a voice note (mp3/m4a/wav) so Przemek transcribes and summarises it
P1 [M]: As the System, I transcribe the voice note (Whisper STT), then AI-transform it into a structured CEO briefing (3-sentence summary, action items with owners and deadlines, key decisions)
P1 [M]: As CEO, I receive the briefing form, review/edit action items, and approve for distribution to team
P2 [D]: As a team member, I receive my assigned action items via email with deadline and link to full briefing

Section 04 -- System Flows [M/D]

Flow A: Marcin -- Lead Qualification (6 Steps) [M]

Trigger [M]: Cron-style scheduler polls SMS gateway inbox. New message detected.

#	Step [M]	Actor [M]	Input [M]	Output [D]	Decision [M]
1	SMS Ingest	System	SMS webhook: from_number, message_body, received_at	SmsLead entity draft created	--
2	ChameleonV2 Lead Form (auto-populated)	System	SmsLead entity fields	Name, Company, Need, Budget extracted from message body	All fields present? If missing: follow-up SMS sent
3	Qualification Rules	System	Lead form fields	Score (rules passed / total). Qualified if score >= 0.66	Qualified: advance to Schedule. Not qualified: SMS decline + Store.
4	Schedule Appointment	System	Calendar API query	appointment_id, appointment_datetime, meeting_link	--
5	SMS Reply	System	Qualification result + appointment details (if qualified)	SMS sent: confirmation with link OR polite decline	--
6	Store SmsLead (deltaPrism)	System	All lead fields + score + result + appointment_id + reply_timestamp	SmsLead entity persisted with run_id	--

Result [M]: Lead fully processed: qualified + appointment booked + confirmation sent, OR declined + polite SMS sent. All data stored in deltaPrism. Fully autonomous -- no human gate.

Flow B: Irena -- Document Lifecycle (6 Steps) [M]

Trigger [M]: Operator uploads document via ChameleonV2 Document Upload form.

#	Step [M]	Actor [M]	Input [M]	Output [D]	Decision [M]
1	Document Upload (ChameleonV2)	Operator	Document file + optional title + sender name	Document Reference BDT stored	--
2	Atlas OCR Extraction	System	Document image/PDF	extracted_text, document_type, confidence_score (0.0-1.0)	--
3	Categorisation Rules	System	OCR output + extracted text	Category: Invoice/Contract/Certificate/Correspondence/Other	High confidence (>= 0.75 AND OCR >= 0.8): Store. Low confidence: Reviewer Gate.
4	Reviewer Human Gate (low confidence only)	Reviewer	Document thumbnail, OCR excerpt, category suggestion + confidence %	Confirmed or overridden category + reviewer_user_id	HUMAN GATE -- pipeline pauses until Reviewer decides
5	Store IrenaDocument (deltaPrism)	System	Document ref, extracted text, category, confidence, reviewer decision (if applicable)	IrenaDocument entity persisted	--
6	Email Forward	System	Category + document attachment + OCR summary	Email sent to category mailbox (invoices: accounting@, contracts: legal@, certificates: hr@)	--

Result [M]: Document classified, stored with full OCR data, and forwarded to correct department. Low-confidence documents reviewed by human before routing.

Flow C: Przemek -- CEO Briefing (6 Steps) [M]

Trigger [M]: Operator uploads voice note (mp3/m4a/wav) or voice webhook fires.

#	Step [M]	Actor [M]	Input [M]	Output [D]	Decision [M]
1	Voice Ingest	System	Audio file (mp3/m4a/wav)	VoiceNote entity: file_ref, duration, recorded_by, recorded_at	--
2	Transcription (Whisper STT)	System	Audio file	transcript_text, language, confidence	--
3	AI Transform (Claude)	System	transcript_text	Structured JSON: executive_summary (3 sentences), action_items[{owner_role, deadline, text}], key_decisions[]	--
4	CEO Briefing Form (Human Gate)	CEO	Structured briefing: summary, action items (editable), key decisions	CEO decision: Approve and Send / Request Revision + edits	HUMAN GATE -- pipeline pauses until CEO approves
5	Store CEOBriefing (deltaPrism)	System	Approved briefing with CEO edits, approval timestamp, CEO user ID	CEOBriefing entity persisted	--
6	Action Item Distribution	System	action_items[] with owner_role mappings	Email per action item to mapped team member. BCC to CEO.	--

Result [M]: Voice note transcribed, AI-structured into briefing, CEO-approved, action items distributed to team. Full audit trail in deltaPrism.

D-05 Resolved: Real Claude API key for AI Transform node. Produces genuinely useful summaries. API key stored in pipeline config (encrypted at rest). Both Przemek and ARC pipelines share the same AI Transform node implementation.

Section 05 -- Data Model [M/D]

Entities

Entity [M]	Pipeline [D]	Description [M]	Key Fields [M]	States [D]
SmsLead	Marcin	Inbound SMS lead with qualification data	lead_id (UUID PK), from_number, name, company, need, budget, qualification_score, result (qualified/declined), appointment_id, appointment_datetime, reply_sent_at, run_id	Draft, Qualifying, Qualified, Declined, AppointmentBooked, Complete
IrenaDocument	Irena	Uploaded document with OCR and classification	doc_id (UUID PK), file_ref, title, sender_name, extracted_text, document_type, category, confidence_score, reviewer_decision, reviewer_user_id, forwarded_to, run_id	Uploaded, OCRComplete, Categorised, ReviewPending, Forwarded
VoiceNote	Przemek	Voice recording with transcription	note_id (UUID PK), file_ref, duration, recorded_by, recorded_at, transcript_text, language, run_id	Ingested, Transcribed, Summarised, PendingApproval, Approved, Distributed
CEOBriefing	Przemek	AI-generated briefing from voice note	briefing_id (UUID PK), note_id (FK), executive_summary, action_items[], key_decisions[], ceo_user_id, approved_at, distributed_at	Draft, PendingCEO, Approved, Distributed

Business Rules [M/D]

Pipeline	ID	Rule [M]	On Violation [D]
Marcin	BR-M01	`budget_above_threshold`: Budget >= 5000 PLN	Score reduced; may lead to decline
Marcin	BR-M02	`need_is_in_scope`: Need matches configured scope keyword list	Score reduced
Marcin	BR-M03	`company_not_blacklisted`: Company not in blacklist	Immediate decline regardless of score
Irena	BR-I01	`is_invoice`: Keywords faktura/invoice/VAT number present	Category not Invoice
Irena	BR-I02	`is_contract`: Keywords umowa/zlecenie/strony present	Category not Contract
Irena	BR-I03	`is_certificate`: Keywords certyfikat/zaswiadczenie present	Category not Certificate
Irena	BR-I04	`low_confidence_flag`: max confidence < 0.75 AND OCR confidence < 0.8	Route to Reviewer Human Gate
Przemek	BR-P01	AI Transform must produce structured JSON with all required fields	Pipeline fails; operator notified

Section 06 -- Screens [D]

Screen [D]	Pipeline	Actor [M]	Key Elements [D]
Lead List View	Marcin	Operator	Table: lead name, company, score, result, appointment link. Read-only.
Lead Override Form	Marcin	Admin	Re-qualify a declined lead. Score override. Reason field.
Document Upload Form	Irena	Operator	Document Reference BDT upload, optional title, sender name.
Category Review Form	Irena	Reviewer	Document thumbnail, OCR excerpt, category suggestion + confidence %, dropdown override.
Voice Note Upload	Przemek	Operator	Audio file upload (mp3/m4a/wav). Optional: title, context note.
CEO Briefing Form	Przemek	CEO	Executive summary (read-only), action items (editable: owner, deadline, text), key decisions. Approve and Send / Request Revision.
Pipeline Canvas	All	All (via Nexus)	Pipeline profile switcher (Marcin/Irena/Przemek). Node graph with status. Run timeline.

Section 07 -- Integrations [M]

System [M]	Pipeline(s) [D]	Direction [M]	Data [M]	Format [D]	Fallback [D]
SMS Gateway (webhook)	Marcin	In / Out	Inbound SMS (from, body, timestamp). Outbound SMS (to, body).	REST webhook POST (inbound); REST API POST (outbound)	Demo: Mock webhook POST to /webhook. Mock send: log payload + display in canvas.
Calendar API (Google Calendar)	Marcin	Out	Create appointment: slot query, appointment creation	REST API (OAuth2)	Demo: Mock returns appointment_id + next available slot.
Atlas OCR Engine	Irena	Internal	Document images in, structured extraction + classification out	Internal pipeline node	OCR failure: flag for manual classification
Whisper STT	Przemek	Internal	Audio file in, transcript text out	Internal pipeline node (API call to Whisper)	Transcription failure: pipeline fails with error
Claude AI (Anthropic API)	Przemek	Internal	Transcript text in, structured briefing JSON out	REST API (API key auth)	AI failure: pipeline fails; operator notified. Real key used (D-05 resolved).
deltaPrism (Entity Store)	All three	Internal	SmsLead, IrenaDocument, VoiceNote, CEOBriefing entities	Internal graph store API	Store failure: pipeline halts. No silent data loss.
Email (SMTP/API)	Irena, Przemek	Out	Document forward (Irena), action item distribution (Przemek)	SMTP or Email API	Demo: Mock logs email content in canvas node detail.

Section 08 -- Infrastructure [D]

Layer	Choice [D]	Reason [D]
Pipeline Engine	Nexus Core (Python rule engine + pipeline orchestrator)	Multi-pipeline support in single org. Pipeline profile switcher. Proven gate patterns.
Frontend Rendering	ChameleonV2 (role-variant form engine)	Single schema per pipeline, role-specific form variants (Operator, Reviewer, CEO).
Entity Store	deltaPrism (graph database)	Three entity classes in one store. Audit trail. Cross-pipeline queries.
OCR	Atlas OCR (Irena pipeline)	Document type classifier pre-trained on Invoice/Contract/Certificate/Correspondence.
AI / NLP	Whisper STT + Claude AI Transform (Przemek pipeline)	Real AI producing genuinely useful output. Not mocked.
Hosting	Containerized (Docker) -- client-deployed	ENDO as managed product: UV delivers pipeline definitions, client hosts Nexus instance.
Auth	RBAC (5 roles: Admin, Operator, Reviewer, CEO, System)	Pipeline gates enforce role-based access. Single org, shared user pool.

Section 09 -- Critical Path [D]

#	Item [D]	Depends On	Duration [D]	Blocker Risk [D]
1	Pipeline graph definitions (Marcin, Irena, Przemek)	--	1 week	Low -- Complete
2	Entity schemas (SmsLead, IrenaDocument, VoiceNote, CEOBriefing)	#1	3 days	Low -- Complete
3	Mock SMS/Calendar/Email nodes	#1	2 days	Low -- D-08 resolved
4	ChameleonV2 form variants (6 forms across 3 pipelines)	#2, ChameleonV2 engine	1.5 weeks	MEDIUM -- depends on ChameleonV2 readiness
5	Atlas OCR integration (Irena: 4 document types)	#2, Atlas engine	1 week	MEDIUM -- depends on Atlas model availability
6	Whisper + Claude integration (Przemek)	#2, API keys	3 days	Low -- D-05 resolved (real API key secured)
7	Pipeline profile switcher (multi-pipeline canvas)	#1, Nexus canvas	3 days	MEDIUM -- Nexus canvas feature
8	Autonomous demo (all 3 pipelines)	#3, #4, #5, #6, #7	3 days	Low -- assembly only

Section 10 -- What Is Needed [M]

#	Item [M]	From [M]	Blocking [M]	Status [M]
1	ChameleonV2 engine operational	Nexus Platform team	Yes -- blocks all 6 form variants	In Development
2	Atlas OCR with Invoice/Contract/Certificate classifiers	Nexus Platform team	Yes -- blocks Irena pipeline	Planned
3	deltaPrism entity store operational	Nexus Platform team	Yes -- blocks all entity storage	In Development
4	Claude API key (secured)	UV -- already resolved (D-05)	No -- key secured	Complete
5	Pipeline profile switcher in Nexus canvas	Nexus Platform team	Yes -- blocks multi-pipeline demo UX	Planned
6	Pryzmat Media client sign-off on pipeline definitions	Pryzmat Media	No -- current definitions valid for demo	Open

Section 11 -- Acceptance Criteria [M/D]

Definition of Done [D]

Criterion [D]	Verified By [D]
Marcin: Full 6-step run autonomous in < 10s	Timed demo run with fixture SMS payload
Irena: High-confidence doc processed in < 20s (no gate)	Upload + OCR + categorise + store + forward in timed run
Irena: Low-confidence doc triggers Reviewer gate	Upload doc with ambiguous text; verify gate activates
Przemek: Voice note to CEO briefing in < 30s	Upload audio; verify transcript + AI summary + CEO form
Przemek: CEO gate pauses until approval	Interactive demo: pipeline waits, CEO approves, items distributed
Pipeline profile switcher shows all 3 pipelines	Canvas displays Marcin/Irena/Przemek graphs; switch between them
All entities stored in deltaPrism with audit trail	Query deltaPrism for each entity type; verify fields + timestamps

Demo Performance Targets [M]

Pipeline	Autonomous Run Time	Interactive Gate	Node Count	RBAC Roles Demonstrated
Marcin	< 10s	None (fully autonomous)	6	System
Irena	< 20s (high conf) / < 60s (with gate)	Reviewer (low confidence only)	6	Operator, Reviewer, System
Przemek	< 30s	CEO (approval gate)	6	Operator, CEO, System

Section 12 -- Risks [D]

#	Risk [D]	Likelihood [D]	Impact [D]	Mitigation [D]
R1	ChameleonV2 not ready -- blocks all 6 form variants	Medium	High	Fallback: static HTML forms for demo; ChameleonV2 integration post-demo
R2	Atlas OCR accuracy low on Polish documents (faktura formatting varies)	Medium	Medium	Confidence threshold + Reviewer gate handles ambiguity
R3	Claude API latency spikes degrade Przemek pipeline speed	Low	Low	Timeout + retry; acceptable for demo; cache templates for production
R4	Pipeline profile switcher UX complexity for multi-pipeline display	Medium	Medium	MVP: simple dropdown selector. Advanced: tabbed canvas view

Decisions Resolved [D]

ID	Decision	Resolution
D-05	AI Transform -- Real Claude API Key	RESOLVED 2026-04-18: Real Claude API key. Produces genuinely useful summaries. Encrypted at rest. Shared between Przemek and ARC pipelines.
D-06	ENDO Org Structure	RESOLVED 2026-04-18: Single Nexus org (Pryzmat Media). All 3 pipelines in one org. 5 RBAC users seeded. Pipeline profile switcher for multi-pipeline demo.
D-08	Mock SMS/Calendar Nodes	RESOLVED: Mock webhook trigger (manual POST). Mock SMS send (log + display). Mock calendar (return fake slot). No real Twilio/Google Calendar required for demo.

Section 13 -- Cooperator Brief [D]

Field	Value
Scope [D]	Three independent AI employee pipelines in one Nexus org: Marcin (6-step SMS lead qualification + appointment booking), Irena (6-step document OCR + categorisation + routing with optional Reviewer gate), Przemek (6-step voice transcription + AI briefing + CEO approval + action item distribution).
Stack [D]	Nexus pipeline engine (Python), ChameleonV2 forms (6 variants), Atlas OCR, deltaPrism (4 entity classes), Whisper STT, Claude AI Transform, Mock SMS/Calendar/Email nodes.
Timeline [D]	Pipeline definitions: Complete. Entity schemas: Complete. Demo: ~3 weeks from ChameleonV2 + Atlas + profile switcher readiness.
Dependencies [D]	ChameleonV2 (forms), Atlas OCR (Irena), deltaPrism (storage), Pipeline Profile Switcher (canvas UX). All internal Nexus platform dependencies.
Interfaces [D]	Receives: SMS webhooks, document uploads, voice note uploads. Delivers: SMS replies, appointment bookings, categorised email forwards, CEO briefings, action item emails.
Quality Bar [D]	3 pipelines running independently. 18 nodes total. 2 Human Gates (Irena Reviewer, Przemek CEO). Full autonomous + interactive demo modes. All entities auditable in deltaPrism.
Handover [D]	Pipeline definitions (3 graphs) + entity schemas (4) + ChameleonV2 form schemas (6) + demo fixture data + mock node specs + Claude prompt templates + demo runbook.

Use Case: ENDO AI -- Three AI Employee Pipelines