SpendWise — Assignment 2
Technical Scope | SE 638 Software Project Management
System context
Architecture
Dependency map
Compliance
Extra credit ✦
System context diagram
Boundaries of the SpendWise recommendation engine, its users, and external integrations.
System boundary Recommendation Engine SpendWise core system SpendWise User purchase intent recommendation Bank APIs Plaid / Finicity transaction data Push service FCM / APNs Payment processor alert trigger installment plan Analytics Segment / Mixpanel Affiliate partners Financial products engagement events referral trigger CFPB compliance boundary
High-level architecture diagram
Three-layer modular architecture. Business logic: Cash Flow Analysis → Recommendation Engine → Compliance / Output Filter → Feedback Loop → ML Scoring Engine (Project 2).
Presentation layer Mobile app UI iOS / Android REST API gateway Auth + routing Notification service FCM / APNs Business logic layer Cash Flow Analysis Runs first Budget eval Recommendation Engine Core microservice Pay/split/postpone Compliance / Output Filter CFPB gate Wording check Feedback Loop Learning signal ML Scoring Engine Project 2 Personalization Data access layer Transaction DB User spend history Recommendation store Outcome log User profile store Budget + consent
Dependency map — module level
Precedence diagramming (S2S, F2S, F2F) at module level — consistent with how Google runs large programs.
Primary delivery chain
Bank API Integration ──F2S──▶ Cash Flow Analysis ──S2S──▶ Recommendation Engine ──F2S──▶ Compliance / Output Filter ──F2S──▶ Notification Delivery
FromTypeToRiskRationale
Bank API IntegrationF2SCash Flow AnalysisHighCannot run without a verified real-time data feed
Cash Flow AnalysisS2SRecommendation EngineMediumBoth run concurrently once analysis service starts
Recommendation EngineF2SCompliance / Output FilterHighFilter must check wording before any output reaches users
Compliance / Output FilterF2SNotification DeliveryLowOnly approved CFPB-compliant content can be delivered
Project 2 dependencies
Feedback Loop / Recommendation Store ──F2F──▶ ML Scoring Engine  |  Feedback Loop ──S2S──▶ Recommendation Engine
FromTypeToRiskRationale
Feedback Loop / Recommendation StoreF2FML Scoring Engine (Project 2)MediumML cannot complete until outcome history is fully collected
Feedback LoopS2SRecommendation EngineMediumFeedback begins capturing signals as engine starts delivering
Internal dependencies
Transaction DB ──F2S──▶ Cash Flow Analysis ──F2S──▶ Recommendation Engine ──F2S──▶ Compliance / Output Filter ──F2S──▶ Notification Delivery
User Profile Store ──F2S──▶ Recommendation Engine
Recommendation Engine ──F2S──▶ Recommendation Store ──F2S──▶ Feedback Loop ──F2F──▶ ML Scoring Engine
Internal dependencyTypeRiskNote
Transaction DBInternal data storeMediumMust provide transaction and balance history before Cash Flow Analysis starts
User Profile StoreInternal data storeMediumMust provide budget, consent, and preference data before Recommendation Engine runs
Cash Flow AnalysisInternal moduleHighMust complete affordability evaluation before Recommendation Engine decides pay / split / postpone
Recommendation EngineInternal moduleHighMust generate output before Compliance / Output Filter checks the wording
Compliance / Output FilterInternal controlHighMust approve suggestion-only wording before Notification Delivery or mobile output
Recommendation StoreInternal data storeMediumMust log generated and approved recommendations for traceability and future feedback analysis
Feedback LoopInternal moduleMediumMust capture user actions and outcomes after recommendations are delivered
ML Scoring EngineInternal — Project 2MediumNeeds Project 1 feedback and outcome history before ML personalization can be trained
External dependencies
DependencyTypeRiskMitigation
Plaid / Finicity APIExternal vendorMediumConfirm rate limits and SLA before Cash Flow Analysis development begins
FCM / APNsExternal vendorLowStable standard integration
CFPB compliance reviewRegulatoryHighLegal sign-off required before Compliance / Output Filter content is finalised
Stripe / BraintreeExternal vendorMediumValidate installment plan API in discovery sprint
ML training dataInternal — Project 2MediumProject 1 Feedback Loop must accumulate sufficient outcome history first
Regulatory and compliance considerations
Requirements mapped to system components. HIPAA excluded — not applicable to fintech.
StandardKey requirementsComponents affectedStatus
CFPBSuggestions only — not regulated financial advice. No guarantee language.Recommendation engine, Compliance/Output Filter, notification copy, mobile UILegal review required
PCI-DSSTokenization of payment data, firewall config, secure transmissionPayment processor, Transaction DB, API gatewayVendor handles majority
CCPAUser consent for transaction data, right to deletion, opt-out of sharingUser profile store, consent logging, bank API integrationToS update needed
GDPRData encryption, consent logging, right to be forgotten, data minimisationAll data stores, authentication layer, user profile storeIf EU users onboarded
Inv. Advisers ActSpendWise must not constitute registered investment advice. Framing critical.All user-facing recommendation copyHard constraint
HIPAA deliberately excluded — it covers healthcare data, not financial platforms. Including it would indicate template-copying rather than project-specific analysis.
Advanced concepts — extra credit
Not bolted on — each concept traces directly back to Assignment 1 decisions.
SAFe architectural runway
The modular add-on architecture in Project 1 was designed to avoid blocking Project 2. The recommendation engine microservice is decoupled from the ML scoring engine — Project 2 plugs in without touching Project 1 code. The feedback loop begins collecting outcome data in Project 1 so that by the time Project 2 starts, there is enough training signal. This is what SAFe calls "runway" — the architecture is already prepared for the next horizon before you get there.
Hybrid delivery model
Bank API integration and the compliance layer are planned predictively — requirements are fixed, no iteration needed. The recommendation engine and notification UX are delivered adaptively — early user feedback shapes which trigger conditions and copy perform best. The Compliance / Output Filter is a hard gate on every sprint, ensuring CFPB compliance is part of the definition of done, not a post-sprint review.
DevOps integration planning
(1) Feature flags — recommendations roll out to 5% of users first, churn and engagement measured before full release. (2) CI/CD pipeline — new model versions tested against a holdout set before deployment, zero-downtime releases. (3) Automated compliance testing — recommendation copy is linted against CFPB language rules on every commit. This connects directly to the Assignment 1 success criteria — fast releases measured against churn and conversion targets.
All three concepts trace back to Assignment 1 — the two-project program structure, the adaptive/hybrid lifecycle, and the MOV metrics. This assignment extends the same thinking technically.