System Architecture

System Architecture Overview

Cernio is designed as a multi-tenant SaaS platform with a web-first interface and a mobile companion application. The architecture is composed of five major layers:

    Client Layer
         ↓
  Application Layer
         ↓
AI Orchestration Layer
         ↓
     Data Layer
         ↓
External Data Sources

Each layer has clearly separated responsibilities.

Client Layer

The client layer contains the interfaces used by the exporter.

Web Application

Primary working environment. Main responsibilities:

buyer discovery
company research
decision maker discovery
lead workspace
follow-up management

Technology:

Next.js
React
TypeScript

Mobile Companion App

The mobile app focuses on field workflows. Main responsibilities:

business card scanning
meeting notes
quick lead creation
follow-up reminders

Technology:

React Native
Expo

The mobile app is not intended to replicate the full web interface. Instead, it acts as a field capture tool.

Application Layer

The application layer orchestrates platform logic. Main responsibilities:

Authentication
Organization management
Search orchestration
Lead management
Credit accounting
AI pipeline triggering

Typical request flow:

User request
      ↓
  API Gateway
      ↓
Application Service
      ↓
  AI Pipeline
      ↓
   Database
      ↓
   Response

API Layer

The API layer exposes platform capabilities. Example endpoints:

endpoint	description
POST /search	buyer discovery
GET /companies/{id}	company details
GET /contacts/{company}	decision makers
POST /leads	save lead
PATCH /lead/{id}	update status

AI Orchestration Layer

The AI orchestration layer coordinates different AI tasks. Responsibilities:

model routing
prompt management
pipeline execution
provider fallback
cost optimization

This layer prevents direct calls from the frontend to LLM providers.

AI Provider Strategy

Different AI tasks require different models. Example routing strategy:

task	model
query expansion	Gemini
company analysis	Claude
contact extraction	GPT
web reasoning	Perplexity

This approach optimizes:

cost
latency
reasoning quality

AI Routing Logic

The routing logic determines which model should execute each task. Example logic:

if task == query_expansion:
    use Gemini

if task == company_analysis:
    use Claude

if task == contact_extraction:
    use GPT

Fallback logic:

if provider fails:
    switch to backup provider

Retrieval Service

The retrieval service performs company discovery. Responsibilities:

execute search queries
collect candidate companies
aggregate results
remove duplicates

Retrieval sources include:

search engines
trade directories
company websites
LinkedIn company pages

Enrichment Service

The enrichment service analyzes company data. Responsibilities:

industry classification
distributor identification
sector relevance
company size estimation

This step transforms raw companies into structured company profiles.

Scoring Service

The scoring service evaluates companies using deterministic scoring. Inputs:

industry match
distributor probability
country match
company size

Outputs:

FitScore
priority ranking

Lead Management Service

The lead management service powers the CRM workflow. Responsibilities:

lead creation
status tracking
note storage
follow-up reminders
interaction history

Data Layer

The data layer stores structured platform data. Technology:

Supabase
Postgres

Key features:

row level security
multi-tenant isolation
relational data model

Multi-Tenant SaaS Model

The system supports multiple organizations. Core entities:

organizations
users
organization_users
subscriptions
credit_wallets

This structure allows:

team accounts
billing per organization
shared lead workspace

Database Entity Model

The database includes the following primary entities.

organizations
users
companies
contacts
leads
interactions
search_history

Entity Relationships

Core relationships:

Organization
 └ Users
 └ Leads
 └ Searches

Company
 └ Contacts

Lead
 └ Company
 └ Contact
 └ Interactions

Lead Entity

The lead table stores discovered opportunities. Example structure:

field	description
id	lead id
company_id	linked company
contact_id	linked contact
status	pipeline stage
notes	user notes
follow_up_date	reminder

Contact Entity

Contacts represent decision makers. Fields include:

field	description
name	person name
title	job title
email	contact email
linkedin	LinkedIn profile
confidence	discovery confidence

Search History Entity

Searches must be stored for analytics. Example fields:

field	description
product	searched product
country	target market
queries_generated	search queries
companies_found	results

Caching Strategy

AI discovery is expensive. Caching helps reduce cost. Example cache key:

hash(product + country)

Cache policy: cache result for 30 days Users can trigger: search again to force a new run.

Credit Economy

AI processing consumes credits. Example credit costs:

action	credits
buyer discovery	1
contact reveal	1
deep company analysis	2

Credits ensure cost control.

Observability

The platform must track performance metrics. Monitoring includes:

pipeline latency
search success rate
ranking accuracy
credit usage
system errors

This ensures continuous improvement.

Security Model

Security features include:

row level security
API authentication
organization data isolation
secure AI provider keys

Supabase RLS ensures tenants cannot access each other’s data.

Infrastructure Strategy

Initial infrastructure is lightweight. Example stack:

Next.js (frontend)
Supabase (backend)
Edge functions
AI provider APIs

Later scaling may introduce:

worker queues
vector databases
analytics warehouse

Architecture Summary

The system combines:

AI discovery
buyer engagement workflow
lead memory
data intelligence

into a unified SaaS platform.

Overview

Business

Handbook

Strategy

Design

Reference