Farolito AI

Formerly known as Lighthouse AI; renamed due to existing trademark.

Guiding the way to government benefits. A project for the Out of Pocket AI Hackathon, June 2024 in San Francisco.

Team

• Albert Cai
• Irene Jiang
• Nathan Leung

Limitations

• slow
• demo only runs on max first 6 pages of pdf
• pdf is flaky

Repository Structure

assets

Static assets (pdfs, fonts, etc.)

pdfs

code for converting pdfs to images, writing text to pdfs, sending pdfs to openai

Getting Started

1. pip install -r requirements.txt
2. Set OPENAI_API_KEY in .env (can be at repository root)

Deploy

(replace with your username/container registry)

1. docker build --platform linux/amd64 --tag ghcr.io/nathanhleung/farolito/pdfs:latest . (in pdfs directory)
2. docker push ghcr.io/nathanhleung/farolito/pdfs:latest
3. Set image id in Runpod, set OPENAI_API_KEY environment variable in Runpod

ingest

FastAPI endpoints to read DL and paystub info

Getting Started

1. pip install -r requirements.txt
2. Set OPENAI_API_KEY in .env (can be at repository root)

Deploy

(replace with your username/container registry)

1. docker build --platform linux/amd64 --tag ghcr.io/nathanhleung/farolito/ingest:latest . (in ingest directory)
2. docker push ghcr.io/nathanhleung/farolito/ingest:latest
3. Set image id in Runpod, set OPENAI_API_KEY environment variable in Runpod

frontend

frontend for the app

1. npm install
2. Set NEXT_PUBLIC_INGEST_API_ENDPOINT in .env.local (in frontend directory) — based on the pod ID and exposed port (probably 8000) from Runpod, include protocol, no trailing slash
3. Set NEXT_PUBLIC_PDFS_API_ENDPOINT in .env.local, include protocol, no trailing slash
4. npm run dev

Deploy

1. Set NEXT_PUBLIC_INGEST_API_ENDPOINT, NEXT_PUBLIC_PDFS_API_ENDPOINT, NEXT_PUBLIC_ASSISTANT_WEBSOCKET_ENDPOINT in Vercel
2. git push

test

Files used for testing