The Scenario
You joined an HR tech company six weeks ago as a developer. One of the first things your team asked for is a cleaned skill taxonomy — they've been extracting skills from resumes with an NLP pipeline, and the output is a mess. "ML," "machine learning," "Machine Learning," and "deep learning" are all different values in the current dataset. The Google Sheet has 100 raw skills. You need to normalize them against PDL's skill taxonomy before the dataset can feed the recommendation engine. Your manager mentioned it in standup on Monday. It is now Thursday.
The bad version:
- Write a fuzzy-match deduplication script against a static skill list you downloaded from somewhere on GitHub — it handles the obvious cases but doesn't know PDL's taxonomy at all
- Manually look up each edge case in PDL's API console and copy the canonical name back into the sheet
- Find at row 60 that "deep learning" maps to a different canonical cluster than "ML" in PDL's taxonomy, which is correct, but now your static dedup script is contradicting what PDL says and you have to decide which to trust
You've now spent a full day on a preprocessing step that should have been 10 minutes of tooling.
The Easy Way: One Prompt in SheetXAI
SheetXAI is an AI agent that lives inside your Google Sheet. It reads the raw skills column and uses PDL's skill enrichment endpoint to normalize each entry to a canonical skill name and category.
For each raw skill in column A, enrich it via PDL skill enrichment and write the canonical skill name and related skill category into columns B and C.
What You Get
- Column B populated with PDL's canonical skill name for each raw input
- Column C populated with the skill category from PDL's taxonomy — for example, "Machine Learning" categorized under "Data Science," "Python" under "Programming Languages"
- Rows where PDL cannot confidently classify the skill flagged in column D so you can handle them in the pipeline separately
What If the Data Is Not Quite Ready
The raw skills include multi-skill strings like "Python / R / SQL" that need splitting first
Before enriching, split any multi-skill strings in column A on forward slash, comma, or "and" into separate rows. Then run PDL skill enrichment on each individual skill and write canonical name and category to columns B and C.
You want to collapse synonyms into a single canonical form for the recommendation engine
Enrich each skill in column A via PDL and write the canonical name to column B and category to column C. Then identify any rows in column B that share the same canonical name — mark them as "Synonym of row X" in column D so the pipeline can merge them.
Some rows are job titles that slipped into the skills column — filter them out first
For each entry in column A, attempt PDL skill enrichment. If PDL classifies the entry as a job title rather than a skill, flag it in column D as "Title — remove from skills dataset" and leave columns B and C blank.
Normalize skills, add a proficiency weight column, and output a recommendation-engine-ready format
Enrich each skill in column A via PDL and write canonical name and category to columns B and C. Add a column D weight for use in the recommendation engine: 3 for programming languages, 2 for frameworks and tools, 1 for soft skills and other categories.
One prompt normalizes the taxonomy and adds the weighting schema — the recommendation engine gets a clean input file without a separate preprocessing step.
Try It
Get the 7-day free trial of SheetXAI and open any Google Sheet with a column of raw skill strings from resumes or surveys. Ask it to normalize each skill via PDL and write canonical names and categories back. Then see how to deduplicate a merged contact list using PDL identity resolution or go back to the People Data Labs overview.