Impulse AI

Autonomous MLE agent. Build AI models from data with no code

[Eshan Chordia @eshanchordia]

@spurs 4, @okcthunder 3.

Two weeks ago every source had the Thunder:
@Kalshi: 66% Thunder
@Polymarket: 68% Thunder
Basketball Ref: 68.2% Thunder
@ImpulseAI_: 52% Spurs

We were the only ones picking the Spurs!

[Eshan Chordia @eshanchordia]

Our model gave the @nyknicks a 91% chance to beat the @cavs.

@Kalshi: 70%
@Polymarket: 70%
Basketball Ref: 63.6%
@ESPN BPI: 80%
@ImpulseAI_: 91%

Knicks swept it 4-0. The closest to our call was @ESPN BPI.

[Eshan Chordia @eshanchordia]

The @NBA Conference Finals are here!

@spurs vs @okcthunder:
@Kalshi: 66% Thunder
@Polymarket: 68% Thunder
Basketball Ref: 68% Thunder
@impulseai_ : 52% Spurs (pre-game 1)

We're the outliers again. Let's not count the Thunder out, but we all want the spurs to win ;)

[Eshan Chordia @eshanchordia]

Two weeks ago, every public source had the @DetroitPistons beating the @cavs. Except us!

@Kalshi: 52% Pistons
@Polymarket: 53% Pistons
Basketball Ref: 73.4% Pistons
@ESPN BPI: 80.5% Pistons

@impulseai_ : 70% Cavs

Cavs won 4-3. 30-50pt swing. Our model was right!

[Eshan Chordia @eshanchordia]

Last week our @ImpulseAI_ agent built an @NBA playoff prediction model.

People asked: can it do harder sports?

Hockey is harder. Goalies, overtime, sparse scoring.

So we put it to the test on the @NHL playoffs ↓

[Eshan Chordia @eshanchordia]

Hockey is harder to predict than basketball. Parity by design, just like the @NBA, plus goalies, overtime, sparse scoring.

We gave the @ImpulseAI_ MLE agent 17k @NHL games and a prompt. Asked it to handle the rest.

It did.

[Eshan Chordia @eshanchordia]

Every public NHL model has the @Avalanche favored over the @mnwild.

Moneypuck: 74%
HockeyStats: 78%
@Kalshi: 79%
@Polymarket: 80%
Our @ImpulseAI_ model: 83%

Not a fight about who wins. A fight about how decisively.

[Eshan Chordia @eshanchordia]

3 minutes of our @ImpulseAI_ MLE agent taking 25 years of @NBA data and building the model behind our playoff prediction app.

No code. No data scientist.

The wild part: the dataset is only 75% complete. Most ML tools break here. The agent treats the gaps as signal and ships.

[Eshan Chordia @eshanchordia]

Yesterday we showed how our @ImpulseAI_ @NBA playoff model compares to @Kalshi, @Polymarket, Basketball Reference, and @ESPN.

The most common question back: how did you actually build the models?

Honest answer: we didn't.

[Eshan Chordia @eshanchordia]

Every public source has the @DetroitPistons beating the @cavs:

@Kalshi: 52% Pistons
@Polymarket: 53% Pistons
Basketball Reference: 73.4% Pistons
@ESPN BPI: 80.5% Pistons
Our model says 70% Cavs

A 30-50pt swing. Either we're wrong or seeing something nobody else is

[Eshan Chordia @eshanchordia]

🥳 I'm excited to to share that @impulseai_ has been accepted into the Apollo accelerator program, backed by @0G_labs and @theBBFund, with support from advisors from @StanfordSBA!

Out of hundreds of applicants, only 10 teams were selected. Pretty cool to be one of them! I’m especially excited to spend the next few months around strong mentors, technical builders, and founders.

Excited to build 🫡

[Eshan Chordia @eshanchordia]

We just launched the next generation of our MLE agent at @impulseai_ 😁

Most people don’t want “an ML model.”

They want to know what is likely to happen next, and what to do about it.

[Tyler Postle @PostleTyler]

"Pivoting feels like a failure."

@eshanchordia pivoted @impulseai_ in October, launched a new product in February, and is shipping the next generation in weeks.

The teams moving fastest right now aren't the ones who got it right the first time.

More on Built to Ship.

[Eshan Chordia @eshanchordia]

We just launched the redesign of our website! The new website is soooo much better than our old 😅

The main goal was to make the product much easier to understand. @impulseai_ lets product managers, data owners, analysts, and operators build production-grade predictive models without needing to wait on a data science backlog. You describe what you want in plain English, upload your data, and get a model you can use for forecasting, churn, fraud, inventory, or other business decisions.

Check out the website and let me know what you think 👇

https://t.co/f1xM9LCQpz

[Eshan Chordia @eshanchordia]

We were able to show a real time demo of our agent training ML models for different use cases! If you're interested in learning more DM me, or check us out at: https://t.co/VXo9R1a8tu

[Eshan Chordia @eshanchordia]

We sponsored DataHacks 2026 at @UCSanDiego and it was awesome!

Huge shoutout to the organizers and volunteers. Super well run, thoughtful logistics, and just a really fun event overall. Great working with Ansh Bhatnagar and Tijil Chhabra, both dudes are phenomenal.

What stood out most was the students and the projects. So much energy in the room. Anyone who thinks Gen Z is lazy has no idea what they’re talking about.

I was blown away by how students used @impulseai_ :
- predicting data center buildout
- tools for better hiking
- lowering carbon emissions by improving recycling

Left feeling very optimistic about the future!

[Eshan Chordia @eshanchordia]

Super excited to be sponsoring DataHacks 2026 at @UCSanDiego hosted by the Data Science Student Society!

If you're a student participating, come check out our workshop on how to use @impulseai_ ! We'll cover:

- How to use Impulse end-to-end from data to model deployment
- How to generate an API key and use the model endpoint
- How to understand if the model is working for you

Really looking forward to all the different apps people build!

[PL_Genesis @PL__Genesis]

ICYMI: Winners of the PL_Genesis Frontiers of Collaboration Hackathon was announced last week! 👀

Here's a recap 👇

[PL_Genesis @PL__Genesis]

Building 'Autonomous ML for Every App' sponsored challenge by @impulseai_ winners are:

1) Sensus AI
2) TrustGraph
3) Ember

[Tyler Postle @PostleTyler]

Eshan's honest answer on how Impulse AI monitors complex agents in production:

"Right now we're running scripts manually."

That's where most teams actually are. Somewhere between "we have evals" and "we have a repeatable measurement system."

@eshanchordia @impulseai_