Revmaker / talk-to-owner-manuals

Owner manual NLP API w/ light front end

Talk to Owner Manuals

Local test:

python3 app.py

API / Frontend

Running python3 app.py starts up a server that has the route /api, which returns JSON. It will also serve static files from /, /static, and /templates. The frontend logic is in static/js/main.js - it's pretty straightforward.

The server is a Flask server on port 8000 (configurable with a PORT environment variable). The debug param is also configurable, eg, DEBUG=True PORT=5050 python3 app.py will start the server in debug mode on port 5050.

Ideas

• Use PDFjs so we can eagerly fetch the PDFs, and just render them when the user presses the button - it will seem much faster. We will start to fetch on any dropdown change, but throttled so we don't go too crazy
• NEXT button to jump to next most similar result - this will also benefit from PDFjs, since we don't need to reload the whole doc, we can just move pages
• Once we are comfortable deleting the OLD app, then we can also delete the directories /NLPModels and /TextDBs
• Make sure all required packages are in a file such as requirements.txt
• Clean up the database situation - ideally should connect to MySQL instance with all MMY combos
• Deploy on AWS - this will involve putting the app behind gunicorn... may as well Dockerize it...
• make a /voice endpoint, that returns something like "On page x of the manual_type manual, it says ... "
• Improve embedding - this would mean (possibly) leaving gensim, which sucks as it has a nice API. So it's only worth it for a big improvement
• Do a real frontend setup with npm and some build tools

Action items

Deploy to AWS / Voice

1. Dockerize
2. Deploy to AWS Elastic Beanstalk
3. Make Lambda function that proxies Alexa requests

Make work for all manuals

1. Make the server work for many models - HOW?
2. Create those models and store in S3
3. Make the server fetch models from S3 instead of locally
4. Move SQL info to Postgres