Does Your Data Platform Analyze?
We reverse-engineered Uber, Lyft, and Yellow Taxi pricing formulas from 27 million trips
Most data platforms stop at storage and querying. Import your data, run some SQL, get a table back. Useful -- but limited. You know what happened. You don't know why.
HULDRA changes that.
What HULDRA Does
HULDRA is Chaprola's built-in nonlinear optimizer. An AI agent proposes a mathematical model -- any relationship it suspects might exist in the data -- and HULDRA finds the parameter values that best fit. One API call. One endpoint. Results in under a second.
The agent doesn't need to know the answer in advance. It reads the schema, forms a hypothesis, writes a short objective function, and calls /optimize. HULDRA runs the math and returns the fitted parameters, residuals, and iteration count.
The Test: 27 Million Rideshare Trips
We pointed an agent at the NYC TLC rideshare dataset -- 27 million Uber, Lyft, and Yellow Taxi trips. The agent imported the data into Chaprola, then asked a simple question: what's the actual pricing formula for each service?
Here's what it found.
Yellow Taxi: fare = $3.19/mile - $0.23/passenger + $7.60 189 records sampled. 72 iterations. 1.3 seconds.
Uber: fare = $2.52/mile + $0.47/min + $5.52 100 records sampled. 14 iterations. 0.15 seconds.
Lyft: fare = $1.88/mile + $0.55/min + $1.94 63 records sampled. 32 iterations. 0.28 seconds.
The differences tell a story. Uber charges more per mile but less per minute -- it penalizes distance. Lyft charges less per mile but more per minute -- it penalizes slow trips. Yellow Taxi ignores time entirely and charges a negative passenger coefficient, meaning solo riders pay more per trip than groups splitting a fare.
None of this was documented anywhere. The agent reverse-engineered it from raw trip data.
Demand Curve
The agent then fit an hourly demand model across the full dataset. Two peaks emerged:
- Morning rush: 8:04 AM -- sharp, roughly one-hour window
- Evening rush: 3:42 PM -- broader, about two and a half hours
- Baseline: 64,256 trips per hour
24 records. 61 iterations. 0.48 seconds.
The evening peak is earlier than most people assume. That's not an opinion -- it's a fitted parameter from 27 million data points.
From "What Happened" to "Why It Happened"
Standard analytics tells you that Uber revenue was $X last month. HULDRA tells you how Uber prices its rides -- the actual formula, derived from data, not documentation.
Every dataset imported into Chaprola is one prompt away from this kind of analysis. The agent writes the hypothesis. HULDRA tests it. No notebooks, no data science team, no weeks of work. One API call.
The HULDRA results from this analysis are live on the Big Data case study.