U
AI/LLM Engineer for Travel Itinerary Optimization
Upwork
, US
Contractor
Remote
$25
- 35
Posted 2 weeks ago
Job Description
About the Project
We're building a travel web app that generates personalized itineraries using LLMs. Our system pulls from a curated local database of places (restaurants, attractions, experiences) combined with real-time web data to create day-by-day travel plans.
The core challenge: We need to make itinerary generation faster and higher quality.
---
What We Need Help With
- Speed Optimization
- Reduce time-to-first-output (currently targeting sub-3s)
- Optimize streaming architecture for real-time day-by-day delivery
- Evaluate and benchmark different models (GPT5, GPT5Mini, GPT5 Nano, or others)
- Prompt engineering for faster inference without quality loss
- Quality Refinement
- Improve logical flow of generated itineraries (meal timing, geographic clustering, realistic pacing)
- Reduce hallucinations and ensure factual accuracy from our POI database
- Better personalization based on user preferences (interests, budget, travel style)
- Architecture & Pipeline
- RAG optimization for our local place database
- Hybrid approach: deterministic pre-filtering + LLM narrative generation
- Token efficiency strategies (we've reduced from 50k to ~1.5k tokens per request)
- Caching strategies for repeated patterns
- Keeping Data Fresh
- Strategies for incorporating real-time/seasonal information
- Web search integration for up-to-date recommendations
- Handling dynamic data (events, openings, closures)
---
Our Current Stack
Supabase (Postgres) for POI database
Edge Functions (Deno/TypeScript)
OpenAI API (direct integration)
Streaming SSE architecture
React frontend
---
Ideal Candidate
- Deep experience with LLM optimization (prompt engineering, model selection, latency tuning)
- Familiarity with RAG architectures and hybrid AI systems
- Experience with streaming responses and real-time AI applications
- Understanding of structured output (JSON mode, function calling)
- Bonus: Experience with travel/location-based applications
- To Apply
Please share:
Examples of LLM optimization work you've done (latency improvements, quality gains)
Your approach to balancing speed vs. quality in generative AI
Any experience with travel, maps, or location-based AI systems
Your availability and preferred engagement model
We're building a travel web app that generates personalized itineraries using LLMs. Our system pulls from a curated local database of places (restaurants, attractions, experiences) combined with real-time web data to create day-by-day travel plans.
The core challenge: We need to make itinerary generation faster and higher quality.
---
What We Need Help With
- Speed Optimization
- Reduce time-to-first-output (currently targeting sub-3s)
- Optimize streaming architecture for real-time day-by-day delivery
- Evaluate and benchmark different models (GPT5, GPT5Mini, GPT5 Nano, or others)
- Prompt engineering for faster inference without quality loss
- Quality Refinement
- Improve logical flow of generated itineraries (meal timing, geographic clustering, realistic pacing)
- Reduce hallucinations and ensure factual accuracy from our POI database
- Better personalization based on user preferences (interests, budget, travel style)
- Architecture & Pipeline
- RAG optimization for our local place database
- Hybrid approach: deterministic pre-filtering + LLM narrative generation
- Token efficiency strategies (we've reduced from 50k to ~1.5k tokens per request)
- Caching strategies for repeated patterns
- Keeping Data Fresh
- Strategies for incorporating real-time/seasonal information
- Web search integration for up-to-date recommendations
- Handling dynamic data (events, openings, closures)
---
Our Current Stack
Supabase (Postgres) for POI database
Edge Functions (Deno/TypeScript)
OpenAI API (direct integration)
Streaming SSE architecture
React frontend
---
Ideal Candidate
- Deep experience with LLM optimization (prompt engineering, model selection, latency tuning)
- Familiarity with RAG architectures and hybrid AI systems
- Experience with streaming responses and real-time AI applications
- Understanding of structured output (JSON mode, function calling)
- Bonus: Experience with travel/location-based applications
- To Apply
Please share:
Examples of LLM optimization work you've done (latency improvements, quality gains)
Your approach to balancing speed vs. quality in generative AI
Any experience with travel, maps, or location-based AI systems
Your availability and preferred engagement model
Interested in this role?
Don't miss out! Click below to view the full application on the company's website.
Apply on Company WebsiteRelated Job Opportunities
U
T
U
We Are Hiring UGC English with New Zealand accent, Maori speakers! No experience necessary!
Upwork
,
Contractor
Remote
A
Care Navigator- Remote Schedule Available (Must reside in IL, WI, IA)
ascend.mysmartpros
, US
Full-time
Remote
A
Sales Development Representative (US-based candidates only)
ascend.mysmartpros
, US
Full-time
Remote
