Danceiny/gotry

What it solves

GoTry solves the problem of unreliable travel planning by separating language tasks from mathematical ones. While typical AI assistants often hallucinate prices or feasibility, GoTry ensures that every itinerary is backed by verifiable data, preventing the user from receiving impossible or inaccurate travel plans.

How it works

The project uses a hybrid architecture: a Large Language Model (LLM) handles the conversational interview and translates user intent, while a deterministic TypeScript kernel performs the actual calculations. Every number provided to the user is tagged with a specific source, and if a trip is not feasible, the system explains why and suggests the exact changes needed to make it possible. Unverifiable claims are blocked by a "fact gate" to ensure only trusted information is presented.

Who it’s for

GoTry is designed for travelers who want reliable, evidence-based trip planning and for agent builders interested in creating AI systems that prioritize deterministic accuracy over model-generated guesswork.

Highlights

• Evidence-first design: Every piece of data is traced to a source, and unverifiable claims are never presented as facts.
• Deterministic logic: The LLM does not perform arithmetic or feasibility verdicts; these are handled by code to ensure accuracy.
• Wish pool: If a trip isn't possible under your current constraints, the system saves the destination and the specific conditions required to make it feasible later.
• Privacy-focused: The system uses a read-only approach to browser data and does not store user passwords or sensitive credentials.