About the role
Loger uses language and vision models to read supplier catalogs and fill in product data, and it asks the merchant when it isn't sure. You would make that more accurate, faster and cheaper, and build the tests that tell us whether a change made things better.
What you'll do
- Improve extraction from PDFs, screenshots and spreadsheets: column mapping, variants, prices and images.
- Build and run evaluations on real supplier files, and decide changes from the numbers.
- Decide when the model should make a call and when it should ask the merchant.
- Manage model cost, speed and rate limits, and move us to new models safely.
- Work with the engineers on the import pipeline and the review screen merchants see.
What we're looking for
- 3+ years of software engineering, including at least 1 year shipping an LLM feature to real users.
- Experience with structured outputs, tool calling and handling model failures.
- You have built an evaluation set and used it to compare prompts or models.
- Strong TypeScript or Python. Our code is TypeScript.
- Comfortable working with messy real data: CSV, Excel, PDF text and tables.
- Clear written English.
Nice to have
- Vision models, OCR or document layout analysis.
- E-commerce product data: variants, SKUs, barcodes, price lists.
- Cutting model cost without losing quality.