Project Overview
doko.ai is an llm-based travel assistant designed to provide users with up to date and relevent information about Japan travel.
Project Timeline
add more tool uses and routing to enchance agent capbabiilties
moving over to agentic assistant. working on various tool calls along with documents. doko.ai can now look up live weather and provide exchange rate deatails.
busy with school work for fall quarter
deployed to Vercel. tested several questions. while it does work, it is slow because of reasoning
connected FastAPI backend with frontend. using deepseek as llm of choice because of its cheap token costs
started to design and build out UI with Next.js and Tailwind CSS
finetuned Granite 8B model using MLX framework and AWS sagemaker. didn't achieve significant improvemnt in responses so pivoted to using RAG system instead
started reddit posts processing. cleaned, summarized and assembled q and a style pairs for finetuning
started development of doko.ai. initial planning and architecture design. set up Next.js frontend and FastAPI backend infrastructure.
Upcoming
- Add agentic features
- Reindex RAG database with updated information
- Add page for flight prices
- Add dashboard page
click "Home" to return to homepage without loading the animation