🚀 Pravah — LLMOps-Driven AI Travel Planner (Early-Stage Foundation)

Pravah is an early-stage, production-ready foundation for an AI-powered travel planning platform. It combines LLM intelligence, real-time maps, and deep observability to demonstrate how a scalable, reliable AI system can be built and monitored in practice.

This project represents the initial, foundational version of Pravah, intentionally designed as a strong technical baseline rather than a feature-complete product.

---

What Pravah Does (Current Scope)

Travel Planning & Navigation

• User selects source & destination using Google Places Autocomplete
• Visualizes real driving routes on Google Maps
• Displays distance & ETA
• Generates AI-recommended intermediate stops (food, nature, heritage, rest)
• Places accurate markers directly on the route
• Presents a clean itinerary summary UI

The current focus is correctness, reliability, and observability — not feature overload.

---

AI Intelligence (Gemini)

• Uses Gemini 2.5 Flash for itinerary generation
• Structured JSON-only responses for reliability
• Intelligent stop placement logic (semantic match + fallback)
• Graceful error handling and retries

This layer is intentionally instrumented for LLMOps observability, making it suitable for real-world AI deployment scenarios.

---

Observability-First Architecture (Datadog)

Pravah is designed with production observability as a first-class concern, even at this early stage.

Metrics Emitted

Backend & API

• pravah.api.request.count
• pravah.api.request.latency_ms
• pravah.api.request.error
• pravah.journey.started
• pravah.journey.completed

LLM (Gemini)

• pravah.llm.request.count
• pravah.llm.latency_ms
• pravah.llm.response.success
• pravah.llm.response.failure

AI Output Quality

• pravah.ai.itinerary.stop.count
• pravah.ai.fallback.used

---

Detection Rules Implemented

• High API error rate
• High LLM latency
• LLM agent error spikes
• Silent failures (no successful LLM calls)
• Integration health monitoring

Each alert is actionable and tagged by environment and service.

---

Datadog Dashboards

The dashboards surface:

• API performance and reliability
• Journey funnel (start → completion)
• LLM latency and success/failure
• AI output behavior
• Container-level CPU metrics (Docker)

These dashboards reflect how early AI platforms should be monitored in production.

---

Architecture Overview

React Frontend (Vite)
        |
        v
FastAPI Backend (Python)
        |
        +--> Google Maps APIs
        |
        +--> Gemini LLM
        |
        +--> Datadog Metrics & Monitoring

All services are orchestrated via Docker Compose.

---

Dockerized Deployment

Prerequisites

• Docker Desktop
• Datadog API key
• Google Maps API key
• Gemini API key

---

Environment Setup (.env)

Create a .env file locally (never committed):

# Datadog
DD_API_KEY=your_datadog_api_key
DD_SITE=us5.datadoghq.com

# Gemini
GEMINI_API_KEY=your_gemini_api_key

# Google Maps
VITE_GOOGLE_MAPS_API_KEY=your_google_maps_key

---

Run the Application

docker-compose up --build

---

Access

Service	URL
Frontend	http://localhost:5173
Backend (FastAPI Docs)	http://localhost:8000/docs
Datadog	https://app.datadoghq.com

---

How to Run This

1. Clone the repository
2. Create a .env file
3. Run docker-compose up --build
4. Use the web UI
5. Observe live metrics and alerts in Datadog

This workflow mirrors real production onboarding.

---

Security & Best Practices

• Secrets stored only in environment variables
• .env excluded via .gitignore
• No hard-coded credentials
• Reproducible Docker-based setup

---

Planned Future Evolution

Pravah is intentionally positioned as a starting point for a larger platform:

• Voice-based trip planning
• Real-time navigation updates
• Personalization and memory
• Cost-aware LLM routing
• Distributed tracing (APM)
• Multi-agent planning and feedback loops

---

Author

Built as a foundational prototype to demonstrate how modern AI systems should be designed, observed, and deployed from the very beginning.