#ArtificialIntelligence #DataScience #SQL #Startup #SaaS
— -
The 3-Day Answer
Here's a scene that plays out at every data-driven company, every week.
A product manager has a question: *"What's our retention rate for users who signed up through the referral program last quarter?"*
Simple question. Should take seconds. Instead, it takes days.
She messages the data team. They're mid-sprint. Her request joins a queue. Two days later, an analyst writes a query against three tables, realizes the referral_source column hasn't been populated consistently since the last migration, rewrites the query with a workaround, runs it, drops the results into a Google Sheet, and sends it over.
By then, the product decision has already been made on gut instinct.
This isn't a people problem. The analyst is talented. The PM asked the right question. The data exists. The problem is the gap between having a question and getting a trusted answer.
We built Donna to close that gap.
— -
What Donna Actually Is
Donna is an AI data intelligence agent. You connect your database — PostgreSQL, MySQL, ClickHouse, Snowflake, or even a CSV file — and ask questions in plain English.
But that description undersells it. Lots of tools generate SQL from natural language. Most of them produce output you'd never share with your manager, let alone put in a board deck.
Donna is different because of what happens before you ask your first question.
— -
The Trust Problem No One Talks About
Every AI data tool demos well. You type a question, it spits out a query, you get a table of numbers. Looks impressive.
But then you try it on your own data. Your schema has 150 tables. Half the column names are abbreviations from 2019. There are three different columns that could mean "revenue." The tool picks one. You have no idea if it picked the right one.
You stare at the output. It could be right. But would you put those numbers in a Slack message to your CEO? Would you make a hiring decision based on them?
Probably not. Because the tool didn't earn your trust. It just generated SQL and hoped for the best.
This is the trust gap — and it's the reason most text-to-SQL tools get tried once and abandoned.
— -
Schema Intelligence: Understanding Before Answering
When you connect a database to Donna, the first thing that happens is a deep schema analysis. Before you type a single question, Donna:
Maps every relationship — not just the foreign keys you declared, but the implicit ones. That `user_id` column in your `orders` table that doesn't have a formal constraint? Donna detects it through naming patterns and type matching. In our analysis of 100+ databases, 59% had more implicit relationships than explicit ones.
Classifies every column by sensitivity — emails, phone numbers, payment info, health records. Each gets a sensitivity label: public, internal, confidential, or restricted. PII is flagged and masked before the LLM ever sees it. We found that 15% of databases had personally identifiable information in columns that weren't being treated as sensitive.
Generates plain-English documentation — for every table and column. What it stores, how it relates to other tables, what the status codes mean. This is the documentation your schema deserves but nobody has time to write.
Produces a health score — a 0–100 rating across performance, data integrity, security, documentation, and consistency. The average score across all databases we've analyzed: 64 out of 100.
This isn't a feature. It's the foundation. When Donna later answers your question about retention rates, she already knows which table tracks signups, what `referral_source` means, which join paths are valid, and which columns contain the dates that matter.
She's not guessing. She studied first.
— -
Ask a Question. Get an Answer You'd Actually Share.
Once the schema is understood, the experience is simple. You type a question:
"What's the month-over-month revenue trend for Q4, broken down by product category?"
Donna translates that into SQL, validates it against her understanding of your schema, runs it, and returns:
- The raw data in a clean table
- An automatic visualization — the right chart type, chosen based on the data shape
- A narrative explanation that puts the numbers in context
- The actual SQL she ran, fully visible, so you can verify the logic yourself
No dashboard to build. No chart configuration. No waiting on anyone.
— -
When Things Go Wrong (They Will)
Real databases are messy. Queries fail. Column types don't match. Tables have edge cases that no schema can fully describe.
Most AI tools surface the error and stop. Donna doesn't.
When a query fails, Donna reads the error, diagnoses the issue, rewrites the SQL, and retries — automatically. She gets up to two fix attempts before giving up. In practice, this self-healing loop resolves the majority of first-attempt failures.
This matters because fragility kills trust faster than anything else. If you ask a question and get an error, you lose confidence in the tool. If the tool fixes its own mistakes transparently, you gain confidence instead.
— -
Compound Questions: The Hard Part
Simple questions are, well, simple. The real test is when someone asks something like:
Compare Q1 and Q2 revenue, show the top 5 products in each quarter, and highlight any products that dropped more than 20%.
That's not one query. That's at least three — and they depend on each other.
Donna handles this through compound reasoning. She breaks the question into sub-tasks, executes them in the right order, and synthesizes one coherent answer with a unified narrative.
This is where the schema intelligence pays off most. Donna doesn't just know your tables — she knows how to coordinate multi-step analysis across them without getting confused about which columns mean what.
— -
Who This Is For
Donna isn't for data engineers who live in SQL all day. They don't need it.
Donna is for the people who have the questions but shouldn't need to write code to get answers:
- Founders tracking unit economics, burn rate, and growth metrics — without waiting on a weekly report
- Product managers measuring feature adoption, funnel conversion, and user segmentation — in real time
- Operations leads monitoring fulfillment, SLAs, and pipeline health — without building dashboards
- Finance teams reconciling revenue, forecasting, and analyzing spend — from the source data, not a copy of a copy of a spreadsheet
The analyst bottleneck isn't a capacity problem. It's an access problem. Donna removes it by making the data directly accessible to the people who need it — with trust built in.
— -
Privacy Isn't a Feature. It's a Requirement.
A quick note on something we take seriously.
Every database credential is Fernet-encrypted at rest and decrypted only in memory during query execution. Sensitive columns are masked before any LLM sees the data. Your data never trains any model. Query results stay within your environment.
We built Donna for companies that care about their data. That means the AI agent handling it needs to care just as much.
— -
Auto-Insights and Dashboards
When Donna finishes analyzing your schema, she doesn't just hand you a health report and wait for questions.
She generates a set of auto-insights — meaningful queries based on what she learned about your data, executed automatically and pinned to a dashboard. The first time you open Donna after connecting a database, you already have charts and narratives waiting for you.
These aren't generic templates. They're generated from your specific schema, your specific tables, your specific business domain. And you can pin any chat result to a dashboard, create new dashboards, and share them with a link.
— -
Try It. Break It. Tell Us.
Donna is live at donna.rutter.ai
The free tier includes database connections, schema analysis, and chat — enough to connect real data and see if this changes how you work. You can also explore public datasets from Kaggle or Hugging Face without connecting anything at all.
No credit card required. No setup. Just connect and ask.
If you're someone who makes decisions from data — or wishes your team could — I'd genuinely love for you to try it and tell us what works and what doesn't. We're building this in public and every piece of feedback shapes what comes next.
Get started at donna.rutter.ai →
— -
If you found this useful, follow Rutter AI for more updates as we build in public. You can also find us on LinkedIn and Twitter/X