AI Travel Planner — Personalized Itineraries with RAG

Built version 1 of an AI-powered travel planning platform that generates personalized itineraries based on a traveler's destination, dates, budget, interests, and preferences. The system combines retrieval-augmented generation (RAG) with structured travel data to produce itineraries that feel hand-crafted, not generic.

Led the full-stack development of v1, with my primary focus on API development and frontend integration. Owned the architecture from data layer through to the user-facing experience.

RAG-Powered Itinerary Engine Designed the core itinerary generation pipeline using LangChain and retrieval-augmented generation. The system retrieves relevant destination data, activity options, and local context, then generates day-by-day itineraries tailored to the user's constraints and preferences.

API Layer Built the entire API surface with FastAPI — itinerary generation endpoints, user preference management, destination search, and budget optimization. Designed for low latency so users get itinerary results in seconds, not minutes.

Frontend Integration Connected the backend to the frontend, ensuring the AI-generated itineraries rendered cleanly with maps, timelines, and activity details. Handled streaming responses so users could see their itinerary being built in real time.

Personalization Logic Built the preference engine that learns from user inputs — travel style (adventure vs. relaxation), budget sensitivity, pace preference, dietary needs, accessibility requirements — and feeds these as context into the RAG pipeline for increasingly relevant results.

Grounding AI outputs in real data was the core challenge. Pure LLM generation hallucinated locations and invented opening hours. RAG with a curated retrieval layer solved this by anchoring every recommendation in verified destination data.

Balancing speed and quality — itinerary generation needed to feel instant. I implemented streaming responses and parallel retrieval to keep the perceived latency under a few seconds while still producing detailed, multi-day plans.

Python, FastAPI, LangChain, RAG (Retrieval-Augmented Generation)