AI-powered travel planning made simple.

Client

Vespucci

Role

Full Stack Developer

Timeline

Launching Soon

Overview.

Vespucci is a multi-tenant SaaS platform for travel agencies — built as a modular Go + React monorepo with 4 deployable services, 11 domain-driven modules, 4 platform services, and 7 shared packages. It handles trip planning, client CRM, double-entry accounting, AI-powered document extraction, supplier management, and reminder workflows — with LLM-powered intelligence via OpenAI and Gemini.

THE CHALLENGE

Travel agencies relied on fragmented tools — spreadsheets for finances, separate CRMs for clients, manual document handling, and no intelligent planning assistance. There was no unified platform that combined trip management, financial ledger, client CRM, document intelligence, and AI-powered content generation in one system.

Architected a full-stack modular monorepo with a Go backend (Gin + GORM + Redis), React 19 frontend with Redux Toolkit + React Query, AI-powered document extraction pipeline (LLM factory with OpenAI/Gemini, schema registry, confidence scoring, 33-step timeline), a double-entry financial ledger with state machine validation, Keycloak 23 for multi-tenant PBAC auth, and production deployment on AWS via Terraform and 7 GitHub Actions CI/CD workflows.

THE SOLUTION

Gallery.

Full platform walkthrough — 11 domain modules covering trip planning, AI document extraction (Foundry), double-entry ledger, client CRM, reminder derivation engine, and two-layer supplier catalog.

Video

Main dashboard with Recharts + Chart.js visualizations, KPI metrics, and quick-action navigation — built with 57+ shadcn/ui components and OKLCH color system.

Trip management with status lifecycle (Draft → Proposal → Open → Final → Closed), reference code auto-generation, multi-traveler support, and planner workspace auto-creation.

Canvas workspace with Tiptap (20+ extensions), CodeMirror for JSON/Markdown, AI-assisted generation via OpenAI/Gemini SSE streaming, resizable split view, and optimistic updates with rollback.

Client CRM with encrypted PII, contact points, passport management with expiry tracking, loyalty memberships, relationship tracking, and full-text Postgres FTS search.

Two-layer supplier catalog — 60+ global reference + organization-specific adoption with detection patterns, match keys (ABN, IATA, domain, tax ID), and gold standard classification.

Admin panel with PBAC permission management (resource:action codes), 4-role invitation system (Owner, Admin, Agent, Viewer), and organization-scoped multi-tenant configuration.

TECHNOLOGY

Tech Stack.

Frontend

React 19.1 + TypeScript 5.8

Core framework

Vite 7.1 (SWC)

Build tooling with hot module replacement

Redux Toolkit + Redux Persist

Client state management (19+ slices)

TanStack React Query 5.90

Server state, caching & infinite queries

React Router 7.9

Client-side routing

57+ shadcn/ui + Radix UI + Ariakit

Accessible UI component suite

Tailwind CSS v4 + SCSS + OKLCH

Styling with perceptually uniform color system

React Hook Form + Zod

Form management & validation

Tiptap (20+ extensions)

Rich text editor — collaboration, drag-handle, emoji, image, math, mentions, code blocks, task lists

CodeMirror 6

JSON and Markdown code editing

Recharts + Chart.js

Data visualization & charts

TanStack Table + TanStack Virtual

Tables with virtual scrolling for large datasets

Framer Motion 12

Page transitions with AnimatePresence + 8 custom keyframe animations

Embla Carousel

Carousel components

react-resizable-panels

Workspace split view layouts

cmdk

Command palette for keyboard-driven navigation

jsPDF + html2canvas + react-to-print

PDF generation and export

Sonner

Toast notifications with importance levels

Backend

Go 1.25

Primary backend language

Gin 1.11

HTTP web framework

GORM 1.31

ORM for PostgreSQL 16 with migrations

Redis 7

Caching & session store

Keycloak 23

OAuth2/OIDC with PKCE, hybrid Session + JWT auth

golang-migrate

Version-controlled database migrations

Swagger / swaggo

OpenAPI documentation auto-generation

AWS SDK v2 (S3, SQS, EC2, SSM)

Cloud service integration

golang-jwt v5

JWT token handling with refresh queue

OpenAI API + Gemini

LLM factory pattern for document extraction and trip planning

golangci-lint v2 (15 linters)

Code quality enforcement

testify + go-sqlmock

Backend testing with mocked database

Infrastructure & DevOps

AWS (EC2, ECR, S3, RDS, SQS, SSM, IAM/OIDC, CloudFront)

Cloud platform

Terraform (ECS, RDS, S3)

Infrastructure as Code

Docker multi-stage builds

Go alpine builder → runtime containers

GitHub Actions (7 workflows)

CI/CD — test, lint, build, deploy, release, validate migrations, validate docs

Semantic Release

Conventional commits, automatic changelog, GitHub releases

LocalStack

Local AWS emulation for development

Vitest + React Testing Library

Frontend testing with V8 code coverage

Sentry

Error monitoring and tracing in production

CAPABILITIES

Key Features.

Planner & Itinerary

Canvas-based visual workspace for itinerary design with three view modes (itinerary-only, planner-only, split)

Segment management — flights, lodging, tours, activities, transfers, trains, buses, cruises

AI-assisted generation via OpenAI/Gemini with real-time SSE streaming

Tiptap rich text editor with 20+ extensions — collaboration, drag-handle, emoji, image, math, mentions, code blocks, task lists, typography

CodeMirror integration for JSON/Markdown editing

Schema.org JSON-LD support for structured activity data

Canvas-to-segment conversion (single and batch) with optimistic updates and rollback

PDF export via jsPDF + html2canvas, resizable panels for split view, diagnostics panel for real-time job tracking

AI Document Intelligence (Foundry)

LLM-powered extraction pipeline: PDF ingestion → classification → schema-guided extraction → segmentation → validation

LLM factory pattern — pluggable OpenAI/Gemini clients with provider-agnostic abstraction

Schema registry with versioned canonical schemas and supplier templates

Confidence scoring on extracted fields with 33-step extraction timeline and real-time polling (2s interval)

SQS async job processing with idempotent workers and retry mechanism (up to 3 attempts)

Extraction audit trail with performance metrics (tokens, latency)

Multi-entity attachment — link documents to trips, clients, etc. with tag-based organization

Finance & Ledger

Double-entry accounting model — receivable, payable, refund, adjustment transactions

Invoice lifecycle: Draft → Issued → Partially Paid → Paid (+ Overdue, Voided) with state machine validation

Client and supplier invoices with line items, tax, and discount support

Multiple payment methods (cash, card, bank transfer, wire, ACH, check) with refund processing

Transaction posting and reversal, gross profit and commission calculation

Currency support (ISO 4217) with FX conversion hooks, configurable invoice numbering per organization

Client CRM (Hive)

Client profiles with encrypted PII (given name, family name) and status management

Contact points (phone, email, fax, social media), multiple addresses with role tags (home, postal, billing, shipping)

Passport management with expiry tracking, loyalty program memberships with tiers

Relationship tracking (lead, partner, child, parent, sibling, friend), emergency contacts

Full-text search with Postgres FTS, audit logging of all changes

Reminders & Suppliers

12+ reminder types — deposits, payments, departures, returns, visa checks, follow-ups, birthdays

Derivation engine — supplier reminders auto-generate client reminders with chain integrity validation and T-3 business day logic

Batch operations (up to 100 items), priority levels, assignment and watch lists, activity feed with comment system

Two-layer supplier catalog: 60+ global reference + organization-specific adoption with detection patterns

Match keys (ABN, IATA, domain, tax ID), gold standard classification, processing priority and confidence thresholds

Auth & Multi-Tenancy

Hybrid Session + JWT (Keycloak OAuth2 + PKCE) with automatic fallback

PBAC (Permission-Based Access Control) with granular permission codes (resource:action)

4-role invitation system: Owner, Admin, Agent, Viewer with magic link passwordless login

CSRF protection, security headers (HSTS, CSP), brute-force protection with rate limiting

Force Keycloak logout via hidden iframe, token refresh with concurrent prevention (promise queue pattern)

Trip Management

Status lifecycle: Draft → Proposal → Open → Final → Closed (+ Cancelled)

Reference code auto-generation with configurable organization format

Multi-traveler support (JSONB), currency and timezone per trip, planner workspace auto-creation

Trip cloning, soft delete with recovery, advanced search/filtering/sorting/pagination

UI/UX & Frontend Architecture

57+ shadcn/ui components with Radix UI + Ariakit accessibility

OKLCH color system for perceptually uniform theming with dark mode via CSS custom properties

Redux Toolkit (19+ slices) + React Query dual state management pattern

Command palette (cmdk), virtual scrolling (TanStack Virtual), Embla Carousel, resizable panels

Error boundaries, 5xx diagnostics page with full error trace capture, module-based logger with env var control

INTELLIGENCE

AI / ML Pipeline.

Vespucci integrates AI at three levels: an LLM-powered document extraction pipeline with schema registry and confidence scoring, a trip planning engine with real-time SSE streaming, and developer tooling that automates PR generation, code review, and progress reporting.

Document Intelligence (Foundry)

OpenAI + Gemini (LLM Factory Pattern)

AI-powered extraction pipeline for travel documents

PDF ingestion → classification → schema-guided extraction → segmentation → validation
LLM factory pattern with pluggable OpenAI/Gemini clients
Schema registry with versioned canonical schemas and supplier templates
Confidence scoring on extracted fields with 33-step extraction timeline
SQS async processing with idempotent workers, retry mechanism (3 attempts), audit trail with token/latency metrics

Trip Planner AI

OpenAI GPT-4o-mini

Real-time streaming content generation for trip planning via SSE

Server-Sent Events for streaming AI responses to the frontend
Context-aware prompts built from canvas content, selected text, and user input
Custom plannerllm abstraction in Go — provider-agnostic (OpenAI + Gemini)

AI Developer Tooling

OpenAI (Node.js scripts)

Automated PR generation, code review, and EOD reporting

AI-powered PR creation analyzing git diffs for Conventional Commit-compliant output
Automated code review with intelligent summarization
End-of-day reports auto-generated from commit history

How It Works

The Foundry module uses an LLM factory pattern to abstract AI providers — documents are ingested, classified, and run through schema-guided extraction with confidence scoring across a 33-step timeline. The trip planner uses a custom plannerllm Go package that streams responses via Server-Sent Events. Developer tooling runs as Node.js scripts analyzing git history to automate PR creation, code review, and progress reporting.

IMPACT

Results.

010Domain ModulesDomain-driven Go modules with clean boundaries

020Shared PackagesConfig, editor, schema, SDKs, UI

030+UI Componentsshadcn/ui + Radix UI + Ariakit

040+Tiptap ExtensionsCollaboration, drag-handle, emoji, math, mentions

050CI/CD WorkflowsGitHub Actions pipelines

060+ADRs & RFCsArchitectural decision records

ENGINEERING

Architecture.

01Monorepo: 4 deployable services, 11 domain modules, 4 platform services, 7 shared packages

02go.work for monorepo module management across services, modules, and packages

03LLM factory pattern — pluggable OpenAI/Gemini extraction with schema registry and confidence scoring

0433-step document extraction timeline with SQS async processing and idempotent workers

05Double-entry financial ledger with state machine validation and transaction reversal

06Reminder derivation engine — supplier → client auto-generation with chain integrity and T-3 business day logic

07Two-layer supplier catalog: 60+ global reference + org-specific adoption with detection patterns and match keys

08Hybrid Session + JWT auth (Keycloak OAuth2 + PKCE) with PBAC permission codes and concurrent refresh prevention

09Multi-tenant data isolation with organization-scoped context via X-Agency-ID header

1057+ shadcn/ui components with OKLCH color system and command palette (cmdk)

11Optimistic UI updates with snapshot + rollback pattern on canvas mutations

1245+ ADRs and RFCs documenting every architectural decision

1325+ directories of documentation: ADRs, RFCs, HLDs, standards

Deployment

AWS EC2 + ECR for containerized Go backend deployment

AWS S3 + CloudFront for frontend static hosting with CDN

AWS RDS (PostgreSQL 16) for managed database

AWS SQS for async job processing (document extraction, notifications)

Terraform IaC for ECS, RDS, S3 provisioning

7 GitHub Actions workflows: test-and-lint, build-deploy-backend, build-deploy-frontend, build-deploy-static-website, semantic-release, validate-migrations, validate-docs-taxonomy

Docker multi-stage builds (Go alpine builder → runtime) with Docker Compose for full local stack

LocalStack for local AWS emulation in development

Previous ProjectA CRM built for insurance sales

Next ProjectAll Projects

Get in Touch