Reference Architectures & Case Studies

AWS Serverless

Serverless Custom ERP for Interior Design Logistics

An enterprise-grade serverless ERP system built to coordinate clients, vendors, and internal operations teams in real time.

Features deep tradeoff analysis on serverless cold-starts, connection pooling, and Cognito authorization patterns.

View Reference Architecture →

n8n Integration

Multi-Tenant Chatwoot CRM Contact Sync Pipeline

Dynamic multi-tenant contact and lead synchronization mapping incoming customer chats directly to EspoCRM or HubSpot.

Features conditional data de-duplication lookups and self-healing recovery routes for stale CRM mappings.

View Reference Architecture →

MuleSoft 3-Tier

Identity Resolution & Registry Integration for US Plasma Donors

A high-security, 3-tier API-led MuleSoft architecture orchestrating donor registration, verification, and medical deferral synchronization.

Features PII de-duplication rules using partial SSN hashes, real-time event listeners, and Salesforce Health Cloud integrations.

View Reference Architecture →

AWS Cloud Architecture Case Study

Serverless Custom ERP for Interior Design Logistics

An enterprise-grade serverless ERP system built to coordinate clients, vendors, and internal operations teams in real time.

Architectural Blueprint

graph TD %% Define classes classDef orange fill:#fff,stroke:#ff9900,stroke-width:2px; classDef blue fill:#fff,stroke:#00a1e0,stroke-width:2px; classDef purple fill:#fff,stroke:#9933ff,stroke-width:2px; classDef red fill:#fff,stroke:#ff3366,stroke-width:2px; classDef green fill:#fff,stroke:#24b07c,stroke-width:2px; classDef grey fill:#fff,stroke:#7f8c8d,stroke-width:2px; User["

Users"] Route53["

Route 53"] Amplify["

Amplify (React)"] APIGW["

API Gateway"] User -->|HTTPS| Route53 Route53 --> Amplify User -->|API Requests| APIGW subgraph VPC [VPC - Private Subnet Isolation] Cognito["

Cognito (AuthN)"] LambdaCore["

Core Lambda (Hono)"] Secrets["

Secrets Manager"] RDSProxy["

RDS Proxy"] Aurora["

Aurora Postgres v2"] EventBridge["

EventBridge"] SQS["

SQS Queue"] LambdaWorker["

Worker Lambda"] S3["

S3 (PDF & Catalogs)"] SNS["

SNS"] SES["

SES (Email)"] APIGW -->|Authenticate| Cognito APIGW -->|Route| LambdaCore LambdaCore --> Secrets LambdaCore --> RDSProxy RDSProxy --> Aurora LambdaCore -->|Heavy Job Event| EventBridge EventBridge --> SQS SQS --> LambdaWorker LambdaWorker --> S3 LambdaWorker --> SNS SNS --> SES end %% Assign classes class Route53,Amplify,APIGW blue; class Cognito,Secrets,SES red; class LambdaCore,LambdaWorker orange; class RDSProxy,Aurora purple; class EventBridge,SQS,S3,SNS green;

Technical Dilemmas & Solutions

Dilemma A

Serverless Latency (Hono vs. NestJS)

Dilema: The initial spec proposed NestJS on AWS Lambda. While NestJS provides clean enterprise organization, its heavy DI container leads to cold start latencies of 1.5 to 3 seconds. For a real-time ERP interface, this lag ruins the user experience.

Decision: We replaced NestJS with Hono running on Node.js. Hono is a zero-dependency, ultra-lightweight framework built specifically for serverless edge/lambda environments.

Solution: Cold starts dropped to under 150ms, offering an instantaneous, smooth experience for clients.

Dilemma B

Rigid Identity vs. Dynamic Permissions

Dilema: Managing complex multi-tenant client hierarchies and granular role-based permissions (RBAC) directly inside Cognito Groups is rigid, hard to audit, and bloats JWT token payloads.

Decision: We use Amazon Cognito strictly for Authentication (verifying identity). All Authorization (permissions logic) is managed entirely within PostgreSQL tables.

Solution: This keeps Cognito clean and secure. We can update client limits or vendor permissions instantly in DB tables without modifying Cognito schemas.

Dilemma C

Database Connection Starvation

Dilema: PostgreSQL opens a separate OS process for every connection. In serverless, Lambda functions scale horizontally. A surge of vendors updating stock would spin up dozens of Lambdas, instantly exhausting database connection limits.

Decision: We deployed Amazon RDS Proxy between the Lambda functions and the Aurora database.

Solution: RDS Proxy maintains a persistent connection pool, multiplexing connection requests from active Lambdas to keep database connection counts stable.

Dilemma D

Cost Control (Aurora Serverless v2)

Dilema: While Aurora Serverless v2 scales automatically based on capacity units (ACUs), inefficient queries or excessive client polling will trigger high capacity scaling, leading to unexpectedly high monthly AWS bills.

Decision: We set strict min/max limits on ACU scaling inside Terraform, and implemented React Query cache layers on the frontend.

Solution: The database safely scales down to its minimum capacity (0.5 ACUs) during off-hours, cutting database operational costs by over 60%.

Dilemma E

Long-Running Workloads

Dilema: Generating 200+ page high-resolution catalogs or parsing massive Excel sheets can easily exceed AWS Lambda's strict 15-minute runtime execution limit.

Decision: We used an EventBridge router and SQS queue to process standard async tasks.

Solution: Decoupling the queue means that if import sizes grow, we can migrate the worker Lambdas to persistent AWS Fargate (ECS) containers in Phase 2 without rewriting core API logic.

n8n Integration & Automation Case Study

Multi-Tenant Chatwoot CRM Contact Sync Pipeline

Dynamic multi-tenant contact and lead synchronization mapping incoming customer chats directly to EspoCRM or HubSpot.

Integration Flow (n8n Workflow)

graph TD classDef orange fill:#fff,stroke:#ff9900,stroke-width:2px; classDef blue fill:#fff,stroke:#00a1e0,stroke-width:2px; classDef purple fill:#fff,stroke:#9933ff,stroke-width:2px; classDef red fill:#fff,stroke:#ff3366,stroke-width:2px; classDef green fill:#fff,stroke:#24b07c,stroke-width:2px; classDef grey fill:#fff,stroke:#7f8c8d,stroke-width:2px; Chatwoot["

Chatwoot Trigger"] Validate["

Validate Webhook"] MapChatwoot["

Map Chatwoot"] GetTenant["

Get Tenant Config"] MapCRM["

Map CRM Settings"] CheckMap["

Check Contact Map"] SwitchCRM["

Switch: CRM Vendor"] Chatwoot --> Validate Validate -->|Valid| MapChatwoot MapChatwoot --> GetTenant GetTenant --> MapCRM MapCRM --> CheckMap CheckMap --> SwitchCRM subgraph HubSpotFlow [HubSpot Branch] HubSpotOps["

HubSpot: No Operation"] end subgraph EspoFlow [EspoCRM Branch] MapEspo["

Map EspoCRM"] CheckLead["

Check: Lead Exists?"] FetchContact["

Get EspoCRM Contact"] CreateLead["

Create EspoCRM Lead"] CheckContact["

Contact Exists?"] UpdateContact["

Update Contact"] CreateContactMap["

Create Contact & Map"] FetchLeadInfo["

Get Lead Info"] LeadCondition["

Is Lead Valid?"] UpdateLead["

Update Lead"] DeleteMapping["

Delete Old Map"] end SwitchCRM -->|HubSpot| HubSpotOps SwitchCRM -->|EspoCRM| MapEspo MapEspo --> CheckLead CheckLead -->|Yes| FetchContact CheckLead -->|No| CreateLead FetchContact --> CheckContact CheckContact -->|Yes| UpdateContact CheckContact -->|No| CreateContactMap CreateLead --> CreateContactMap CreateContactMap --> FetchLeadInfo FetchLeadInfo --> LeadCondition LeadCondition -->|Yes| UpdateLead LeadCondition -->|No| DeleteMapping %% Classes class Chatwoot,Validate,MapChatwoot blue; class GetTenant,MapCRM,CheckMap,SwitchCRM orange; class HubSpotOps red; class MapEspo,CheckLead,FetchContact,CreateLead,CheckContact,UpdateContact,CreateContactMap,FetchLeadInfo,LeadCondition,UpdateLead,DeleteMapping green;

Technical Dilemmas & Solutions

Dilema A

Dynamic Multi-Tenant Routing

Dilema: Managing separate n8n workflows for each tenant CRM would be a maintenance nightmare as client onboarding increases.

Decision: Built a single, parameterized workflow using dynamic lookups (GET: Tenant Settings) and a centralized SWITCH router node to direct payloads based on client CRM configurations.

Solution: A single n8n template workflow supports multiple CRMs and dozens of tenants simultaneously with zero maintenance overhead.

Dilema B

Race Conditions & Deduplication

Dilema: Webhook spikes can trigger duplicate contacts in the CRM if multiple chat messages arrive from the same email before the first contact creation database operation completes.

Decision: Implemented a multi-stage validation check (Check if Lead Exists -> Check if Contact Exists on match).

Solution: Ensures the pipeline updates existing CRM records dynamically and keeps contact mapping tables clean, achieving 100% data integrity.

Dilemma C

Stale ID Mapping Recovery

Dilema: When a sales operator converts a lead into a contact in the CRM manually, old ID references in n8n's lookup table cause sync errors on future messages.

Decision: Added a verification branch (LeadCondition) that checks the CRM record state.

Solution: If a mapping reference is stale, n8n triggers the Delete old mapping endpoint automatically, clearing references and starting a clean sync flow on the next update.

MuleSoft 3-Tier Integration Case Study

Identity Resolution & Registry Integration for US Plasma Donors

A high-security, 3-tier API-led MuleSoft architecture orchestrating donor registration, verification, and medical deferral synchronization.

Architectural Blueprint

Salesforce Health Cloud"] MobileApp["

Mobile App (DAPP)"] ExpAPI["

Donor Experience API"] SFHC --> ExpAPI MobileApp --> ExpAPI end subgraph Process Layer ProcAPI["

Donor Process API"] ExpAPI --> ProcAPI end subgraph System Layer SFHCSys["

SFHC System API"] DMSSys["

DMS Database SAPI"] MobileSys["

Mobile System API"] EmailSys["

Email System API"] ProcAPI --> SFHCSys ProcAPI --> DMSSys ProcAPI --> MobileSys ProcAPI --> EmailSys end subgraph Data & Identity Engines SalesforceHC[(Salesforce HC DB)] OracleDMS[(Oracle DB: Donor DMS)] Data36["

Data360 Identity Engine"] AmazonSES["

Amazon SES"] SFHCSys --> SalesforceHC DMSSys --> OracleDMS OracleDMS --> Data36 EmailSys --> AmazonSES end %% Classes class SFHC,MobileApp,SalesforceHC blue; class ExpAPI,ProcAPI,SFHCSys,DMSSys,MobileSys,EmailSys orange; class OracleDMS,Data36 purple; class AmazonSES green;

Technical Dilemmas & Solutions

Dilemma A

US Compliance & PII-Safe Identity Resolution

Dilemma: US regulations require verifying donor identity against national medical deferral registries. However, strict HIPAA compliance prohibits storing full Social Security Numbers (SSN) within Mobile App or intermediate integration layers.

Decision: Configured a secure identity resolution layer via the DMS System API and Data360. We query records using First/Last Name, DOB, and only the last 4 or 5 digits of the SSN (matched as one-way encrypted hashes).

Solution: Achieved 100% compliance, preventing profile duplication and ensuring deferral records match correctly without storing sensitive PII payloads.

Dilemma B

3-Tier API-Led Connectivity

Dilemma: Creating direct peer-to-peer data pipes between Salesforce Health Cloud, the mobile app, and the legacy Oracle DMS database creates high dependency coupling, making schema updates risky.

Decision: Deployed MuleSoft's 3-tier API architecture: Experience APIs (channel-optimized), Process APIs (business logic), and System APIs (secure adapters).

Solution: Decouples backend systems. We can optimize or upgrade the Oracle DB schema underneath dms-sys-api without impacting mobile app operations.

Dilemma C

Event-Driven Deferral Auditing

Dilemma: If a donor is deferred due to a medical exam at a center, promotional emails must stop immediately. Synchronizing this via daily batch runs leaves a compliance window open, exposing the company to penalties.

Decision: Implemented a real-time Event Listener system API. When a deferral is recorded in DMS, it fires an immediate message queue event routed directly to update Salesforce Health Cloud and Marketing Cloud.

Solution: Promotional messaging is suspended in under 1 second of deferral registration, eliminating compliance gaps.

Enterprise Reference Architectures.

Serverless Custom ERP for Interior Design Logistics

Multi-Tenant Chatwoot CRM Contact Sync Pipeline

Identity Resolution & Registry Integration for US Plasma Donors

Serverless Custom ERP for Interior Design Logistics

Architectural Blueprint

Technical Dilemmas & Solutions

Serverless Latency (Hono vs. NestJS)

Rigid Identity vs. Dynamic Permissions

Database Connection Starvation

Cost Control (Aurora Serverless v2)

Long-Running Workloads

Multi-Tenant Chatwoot CRM Contact Sync Pipeline

Integration Flow (n8n Workflow)

Technical Dilemmas & Solutions

Dynamic Multi-Tenant Routing

Race Conditions & Deduplication

Stale ID Mapping Recovery

Identity Resolution & Registry Integration for US Plasma Donors

Architectural Blueprint

Technical Dilemmas & Solutions

US Compliance & PII-Safe Identity Resolution

3-Tier API-Led Connectivity

Event-Driven Deferral Auditing