What does a month of asking an AI assistant actually look like? I exported my Claude conversation history and ran a small NLP pipeline over just my own prompts — cleaning the text, dropping stopwords and code, then pulling out the words, phrases, and themes I kept coming back to. Everything below is aggregated: word counts, themes, and daily activity, never the raw prompts themselves.

64conversations
272prompts
4074words written
15.0avg words / prompt
1489unique terms

Spanning May 3 – May 28, 2026.

What I talked about most

data bias terms model reading google project glove implemented sentences models embeddings weat social test foundation words gender list pipeline names tests user research work table female shows media include python point missing male dataset upload git content slide news request response following non arts

Themes

The same prompts, sorted into the areas I work in. Each share is the portion of topic-bearing words that fell into that theme.

Data Engineering 22.7%

datapipelinetablemissingvaluesdeletion

NLP & Language Models 21.6%

glovesentenceswordswordtokenssentence

Research & Papers 19.6%

readingresearchslideconceptspaperslides

Machine Learning & Modeling 19.0%

modelmodelsembeddingsdatasetlearningrobust

Bias & Fairness 17.2%

biassocialgenderfemalemaleresponsible

Recurring phrases

The two-word combinations that showed up most — a quick fingerprint of the specific things I was digging into (word-embedding bias tests, GloVe vectors, data pipelines).

male femaleglove wikipediaglove twitterwords completecomplete sentencessentences approximatelyapproximately sentencespleasant unpleasantpython googleimplemented weatfemale termsarts termsnames pleasantunpleasant termsgit pull

Daily activity

Prompts per day across the window.

May 3
May 6
May 10
May 13
May 16
May 21
May 25
May 28

Generated from a personal Claude export with a Python + scikit-learn pipeline (TF-IDF, frequency analysis, curated theme matching). Last run 2026-06-01.