Tuned PredictionsGet started

Predictions trained on your data.

Your data.
Your model.
Your predictions.

Train a model to estimate outcomes, classify information, and score new cases. Use its predictions to make more informed decisions. You choose what happens next.

Build your model See it in action

For individuals, startup founders, and teams.

Will this deal close in 7 days?
From past examples to a new predictionCall notes, deal history and known outcomes train your model. A new deal has an approved budget and a contract in review. The model returns an illustrative 78 percent chance of closing within seven days. New outcomes return for the next training run. This is an illustrative workflow; you review model versions before deployment.PAST EXAMPLESNEW PREDICTIONCall notesDeal historyOutcomeYour modelTrained on your outcomesA new dealBudget approved.Contract in review.Ready to predictProcessing…78%Estimated chance of closingwithin 7 days Train, predict, learnCall notes, deal history and known outcomes feed your model. A new deal has an approved budget and a contract in review. The model returns an illustrative 78 percent chance of closing within seven days. New outcomes return for the next training run. This is an illustrative example; you review model versions before deployment.PAST EXAMPLESCall notesDeal historyOutcomeYour modelA new dealBudget approved.Contract in review.Ready to predictProcessing…78%Estimated chance of closing within 7 daysSample prediction
Illustrative sales example. Sample prediction, not measured performance.

Start with a question you already ask.

Explore a sample prediction

Where does this research note belong?

Your past examples

Notes + your category labels

  1. How participants were selected.Methods
  2. The main result of the study.Findings
  3. Why the sample may not generalize.Limitations
Your modelTrained for this question

A new research note

Participants were randomly assigned to two groups. Each group followed a different study procedure.

MethodsCategory

Suggested category

Using your defined categories

Illustrative tasks and sample outputs, not predictions from a live model. Training currently starts with a supported integration; contact us about other datasets.

01 / Build

Start with a question.

Describe the outcome you care about. Tuned Predictions helps prepare examples from your connected data, then brings training and model versions into one place.

Your goal. A plan for your data.

Choose a source and explain what the model should learn. Review the prepared data and the answers it will train on.

“Use my past opportunities to learn which deals close within seven days.”

See the full workflow

Your training goal

Which deals will close in the next seven days?

Prepared examples
Information at the timeKnown outcome
Budget agreed. Contract in review.Closed
Project paused. No start date.Didn’t close

Review the inputs and outcomes before training.

Connected tools.
Useful training data.

Explore integrations

02 / Evaluate

See what your
model learned.

Test on examples kept out of training. Compare the results with a baseline and inspect the mistakes before choosing how to use your model.

Same test examples. Two model versions.Illustrative
Example of reviewing research-note classifications. These are invented results, not measured model performance.
Research noteExpectedBaselineYour model
How the study was runMethodsFindingsMethods
What the study foundFindingsFindingsFindings
Limits of the evidenceLimitationsLimitationsMethods
See what improved and what still needs work. A tuned model can get things wrong, too.

Results you can inspect.

Review overall performance and results for each question. Keep the dataset and evaluation attached to each experiment.

Compare versions on the same test data. A completed training run is a starting point, not proof of a better model.

Explore the product

03 / Use

Put your model
to work.

Use it for yourself, in a project, or with your team. Hosting a model does not mean publishing it for others.

IN THE APP

Ask your model.

Supply a question and text context, or ask supported questions using fresh records from your connected source.

IN YOUR PRODUCT

Make it part of your app.

Call your deployed model through the API. Get a probability, category, or score you can use in your own product.

IN A WORKFLOW

Handle recurring work.

Generate predictions across records, filter the results, and save a workflow you can run again.

Predictions inform. You decide. Outputs are estimates, not guaranteed outcomes. Review the evidence and use your judgment before acting.

Your model can be for your own use from start to finish. You don’t need to publish it or sell access.

New data.
Your next model version.

Schedule data refreshes and retraining as new outcomes arrive. Keep each experiment’s data and results, then decide which version to use.

New outcomes flow through refresh and retraining into a candidate. Human review separates the candidate from the live model. Observed outcomes provide new examples for the next training run.
New outcomesRefresh data + retrainCandidate modelReview before deploymentLive model

Future outcomes feed the next training run.

Retraining can run on a schedule. You review each version before deploying it.

Trace every prediction.

Every prediction has a trace. Follow it back to the model version, source data, and inputs behind the result.

Build your model

For your own use. For what you’re building.

A model for
your kind of work.

01 / Personal projects

Organize your research.

Explore a model that sorts notes using categories you define.

02 / Startup founders

Add predictions to your app.

Use a trained classifier through the API in a product or workflow.

03 / Teams

Learn from past outcomes.

Train on your history to assess new opportunities or support cases.

Explore all use cases

A few useful answers.

Is this only for businesses?

No. Tuned Predictions is for individuals, founders, developers, and teams with a specific question and examples to learn from. Use a model for your own tasks or build it into a product. Publishing or selling access is optional, and is planned for a future release.

Do I need to be an ML engineer?

You can start by describing your goal. You’ll still review the prepared data and choose how to use the results. Initial training and deployment currently need setup support.

What data can I start with?

The current training workflow starts with Attio, Pylon, or Fireflies. Use examples with known outcomes or labels, and only information available at prediction time. For other sources or personal datasets, contact us about the setup.

Does a prediction guarantee an outcome?

No. Predictions are estimates based on the data and model used. They can be wrong. Use them alongside your judgment and other information; you remain responsible for the actions you take.

How do I know whether training helped?

Run an evaluation on held-out examples. Inspect accuracy, mistakes, and available baseline comparisons. Training doesn’t guarantee an improvement; the results help you decide whether a version is useful.

Can I use a model without selling access?

Yes. Publishing is not part of the required workflow. Use your deployed model in the app or through the API for your own projects. Running an exported model on your own machine is a separate path that isn’t a self-service feature today.

Does retraining replace my model automatically?

No. Scheduled refreshes can prepare fresh data and train another version using saved settings. Evaluation and deployment are separate steps. A new training run does not automatically replace your deployed model.

Start with one question.
Make the model yours.

Build your model