Every team at Wayfair work with software, and for every one has a few error messages that just pop up again and again. Once you've run into them a few times, you know exactly how to solve them - but for someone new, this can be huge productivity blockers. Stack Overflow might have most of the answers, but if the software is Wayfair specific, where do you turn?

What it does

Parses screenshots of errors or directly provided error text and matches those errors against a crowd-sourced database of common errors to tell you exactly what to do when you run into that new weird bug.

Uses a bag of words approach, regex, or bigrams to match to existing error definitions.

Automatically detects the language of a submission to categorize, and provides forms for inputting answers to the problems.

Provides the ability to comment and show off exactly how helpful you've been to the Wayfair community.

How we built it

Leveraged tesseract for OCR, flask for the backend, and React for the frontend.

Challenges we ran into

Surprisingly hard to access file objects directly in javascript. Hard to preseed a social database to drive initial adoption - need to consider some automatic ingest of errors.

Accomplishments that we're proud of

Fully functional product in the scope of the challenge day; zero up front work done beyond general concept. Rapid iteration on frontend and backend tables.

What we learned

Lot about tesseract, file uploads, and performance trade offs of various NLP options.

What's next for Kronos Diagnose

Fix current bugs with UI work flow, clean up presentation of solution, enforce standards on solution format to make it cleaner and more actionable. Add more default matches for code types. Improve filtering, sort, discovery, and matching algorithm.Improve scalability of matching process as # of issues expands.

Begin to automatically seed errors for failures in scheduled jobs and seed them with bounties. Integrate more tightly with the Wayfair rewards system to incentivize answering questions. Expand beyond just errors to common questions, similar to stack overflow.

Share this project: