Voice to Text for Data Scientists
Your brain thinks in statistical models and data patterns, but your reports need plain English. Blurt lets you explain your findings by talking — hold a button, speak your analysis, release. Text appears in your Jupyter notebook, Slack message, or stakeholder report. No more staring at a blinking cursor trying to translate your insights into words. Just say what you found and why it matters.
The Typing Problem
Turning analysis into readable narratives
The numbers are clear to you. The correlation is obvious. But explaining it to a product manager who doesn't know what p-values mean? That's the hard part. You could say it out loud in 30 seconds, but typing out a clear explanation takes 10 minutes of rewording and simplifying. The insight loses urgency while you hunt for the right words.
Markdown cells in Jupyter that stay empty
Every notebook should have explanations between the code cells. You know this. But after building the pipeline, cleaning the data, and running the model, who has energy to type documentation? The markdown cells stay empty. Three months later, you open the notebook and have no idea why you dropped those columns or chose that threshold.
Stakeholder reports that take longer than the analysis
The analysis took two hours. Writing the report for leadership takes three. You know exactly what to say — you could present it verbally right now — but converting that into polished written paragraphs is a different skill. You're a data scientist, not a copywriter. The writing drains you more than the statistics.
Explaining methodology choices in peer reviews
A colleague asks why you used gradient boosting instead of logistic regression. You have a clear answer. But typing out the nuance — the distribution skew, the feature interactions, the precision-recall tradeoff — takes 15 minutes. By the time you finish typing, you've lost momentum on the actual analysis you were doing.
Slack questions interrupt your model training
You're tuning hyperparameters when someone asks 'What was the conversion lift from the last experiment?' You know the answer. But typing it means context switching out of your training loop. Five minutes later, you're back but have forgotten which parameter combination you were about to try.
How It Works
Blurt works everywhere data scientists write — Jupyter notebooks, Slack channels, Notion docs, Tableau dashboards, email. Anywhere you need to explain your findings.
Hold your hotkey
Press your chosen shortcut. A small indicator shows Blurt is listening.
Talk through your findings
Explain the insight naturally. Blurt handles punctuation and capitalization.
Release and done
Your explanation appears at the cursor. No copying, no formatting needed.
Real Scenarios
Writing markdown explanations in Jupyter notebooks
You just finished a data cleaning step that drops rows with missing values in three specific columns. Cursor in the markdown cell above, hold button, say 'Removing rows where customer ID, transaction date, or amount are null. These three fields are required for the cohort analysis. This drops about 2 percent of records, mostly from the legacy system migration in 2019.' Documentation done in 8 seconds. Future you will actually understand this notebook.
Explaining statistical findings to non-technical stakeholders
Marketing wants to know if the campaign worked. Hold button and speak naturally: 'The test group showed a 12 percent lift in conversion rate compared to control. This is statistically significant with a p-value under 0.01. In practical terms, if we roll this out to all users, we'd expect about 3,000 additional conversions per month.' Complex stats translated to business impact in one breath.
Documenting feature engineering decisions
You created a clever feature by combining three raw columns. Hold button, explain your thinking: 'The recency score combines days since last purchase, days since last site visit, and days since last email open. Each component is normalized to 0-1 scale then averaged. Higher scores indicate more engaged users. This single feature improved model AUC by 3 points.' The logic is captured while still fresh in your mind.
Quick Slack updates during long-running jobs
Your model has been training for two hours and a PM asks for a status update. Hold button, say 'Model is at epoch 47 of 100. Validation loss is still decreasing, should be done in about an hour. Early results look promising, seeing 8 percent improvement over baseline.' Update sent in 5 seconds without breaking your monitoring flow.
Writing up experiment results in Notion
The A/B test finished and you need to document results for the team wiki. Instead of typing for 20 minutes, talk through it: 'Experiment ran for 14 days with 50,000 users per variant. Primary metric showed 7 percent improvement. Secondary metrics were neutral. Recommendation is to ship variant B to 100 percent of users.' Report section complete. Move on to the next analysis.
Responding to methodology questions in code reviews
A colleague comments asking why you used random forest instead of XGBoost for this particular problem. Hold button, explain: 'Good question. The dataset has only 2,000 samples and 150 features. XGBoost tends to overfit in high-dimensional low-sample settings. Random forest with max depth limits gave better cross-validation stability. See cells 23 through 26 for the comparison.' Technical response posted without the typing friction.
Creating data dictionary entries
The new dataset needs documentation before others can use it. Hold button and describe each field: 'Customer lifetime value is calculated as total revenue minus returns over the customer's entire history. Null values indicate customers with no purchases. Outliers above 50,000 dollars are capped at 50,000 to prevent model skew.' Data dictionary entries written as fast as you can describe them.
Why data scientists choose Blurt over built-in dictation
| Blurt | macOS Dictation | |
|---|---|---|
| Activation | Single hotkey, instant start | Click microphone icon or 'Hey Siri' |
| Speed | Text appears in under 500ms | 2-3 second delay before transcription |
| Technical vocabulary | Handles stats terms like regression, p-value, coefficient | Often mishears technical terminology |
| Reliability | Works consistently across sessions | Frequently fails or stops listening mid-sentence |
Frequently Asked Questions
Start Typing Faster Today
Free to try — no credit card required
Download Blurt