1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
51 small wins to finish your pathNext question →
You have two weeks for a POC, and the customer's data is messy: scanned PDFs, Excel files with merged cells, and emails. What do you do?
30-second answerSay your answer out loud first, then reveal.
Plan for two weeks
| Days | Focus |
|---|---|
| 1–2 | Data inventory: count by type, sample 50 of each, identify the worst offenders |
| 2–4 | Parsing pipeline per type; manually verify parsed output on samples |
| 4–5 | Build the eval set with SMEs (on parseable data) |
| 6–9 | Core AI task (extraction / Q&A) + iteration against evals |
| 10 | Results: accuracy by data type, data quality findings, recommendation |
Techniques by format
- Scanned PDFs: OCR or a vision LLM on page images; measure the character error rate on samples; handwriting may be out of scope.
- Excel with merged cells / multiple tables: programmatic parsing (openpyxl/pandas) with custom logic per template, or render to an image and use a vision model for unusual layouts; preserve headers.
- Emails: strip signatures, disclaimers and quoted replies; reconstruct threads; extract attachments.
Communicating results: "Typed PDFs: 93% field accuracy. Scanned PDFs: 71%, mostly due to low-resolution scans. Recommendation: rescan at 300 DPI or route scans to human review in the pilot." This turns a messy-data problem into an actionable plan.
Related
You understood something today that you didn't yesterday.