IconicShift

IconicShift is an AI-powered strategy platform for founders. I shipped the path that takes a founder’s uploaded documents (PDF, .docx, .pptx) and turns them into validated, structured data.

The core of it is relatively simple:

1. take the file from the user, send it to S3
2. have backend scan the file, extract text store in DB
3. process text as needed to create the structured outputs you need

But there are many footguns that you need to look out for when creating such a pipeline.

For file uploads we’re looking at zip bombs, RCE, ZipSlip, etc. Luckily there are libraries and solutions to help. For LLM provider integrations you’re looking at evals, token spend gating, model versioning, model deprecations, circuit breakers, fallbacks, etc.

With all of this you’re also looking at GDPR / privacy compliance. If a user deletes their account now you need to look throughout your whole system for PII: delete the files from S3, delete the extracted text from the DB. Ooops, did you forget some PII in a <user input> in a prompt that got logged or stored somewhere?

Lots of footguns. It’s not as simple as “call an API” as much as the “it’s an AI wrapper” pundits would like you to believe.