Hello, I'm Maneshwar, and I'm building LiveReview — a blast-radius aware AI code review built for your business-critical systems. Star us to help devs discover the project, give it a try, and share your feedback to help improve the product.
Your Uber says the driver is 7 minutes away.
They show up in 7 minutes.
That is not a lucky guess, that is one of the more quietly insane systems in consumer tech, and it is worth taking apart.
Step 1: chop the world into segments
Uber does not think about roads, it thinks about road segments.
A single road gets cut into a handful of pieces, and globally Uber is tracking around 100 million of these segments.
Each segment has a number attached to it: how long it takes to cross, right now.
To get from your driver to you, the routing engine finds the fastest path through these segments and adds up the crossing times along the way.
That sum is your ETA.
Simple enough, except that number, "how long it takes to cross this segment", is not a constant.
It is 20 seconds at 2am and 2 minutes at 6pm on a Friday, on the exact same 200 meters of road.
Step 2: don't ask Google, ask your drivers
The obvious move is to buy this traffic data from someone who already maps the whole planet.
Uber doesn't. Uber measures it.
Every driver on the platform is already pinging their location every 4 seconds, because that's what the app needs to do anyway.
That stream is basically free traffic telemetry. Every time a driver crosses a segment, Uber now knows, to the second, how long that segment just took.
Multiply that by every active driver and you get a live traffic sensor network that nobody had to install a single camera for.
Step 3: the part that actually needs machine learning
Here's the catch. Live measurements only tell you what a segment did in the past few minutes.
Your ETA needs to know what it's going to do while your driver is still en route to you, which could be 15 minutes from now.
So in 2022, Uber shipped a deep learning system called DeepETA, and it does not just average recent history.
- If a normally fast road has a fresh accident on it, DeepETA trusts the live signal over the historical pattern.
- If a quiet backstreet has no recent driver on it at all, DeepETA infers its state from the busier roads around it instead of guessing blind.
It refreshes these forecasts for every segment, for the next 3 hours, every few minutes, and answers roughly 2 million forecast requests a second, making it one of the busiest models running inside Uber. (Uber Engineering: DeepETA)
flowchart LR
A[Driver GPS pings<br/>every 4s] --> B[Live segment<br/>crossing times]
B --> C[DeepETA<br/>forecasts 3h ahead]
C --> D[Routing engine sums<br/>segments on your path]
D --> E[Correction model<br/>trained on real trips]
E --> F[The ETA on<br/>your screen]
classDef start fill:#e9ecef,stroke:#6c757d,color:#1a1a1a
classDef chip fill:#5ee6c8,stroke:#1f9c86,color:#1a1a1a
classDef accel fill:#9d8cff,stroke:#5b4bcc,color:#1a1a1a
class A start
class B,D chip
class C,E accel
class F startAnd there's one more pass after all that. The routing engine's segment-summed number goes through a second model, trained on millions of completed real trips, whose only job is to catch and correct the places where the physics-based sum tends to be systematically wrong (a stop sign nobody accounted for, a left turn that always takes longer than it looks).
The payoff
Shipping DeepETA improved long-trip arrival accuracy by 6%.
Uber estimates that alone is worth around $100 million a year in gross bookings, because an ETA people trust is an ETA people don't cancel on.
Next time your ETA ticks down without drama, that's 100 million road segments, a live sensor network made of other people's cars, and a model answering 2 million questions a second so a number on your screen can be boring.
Your team's attention is limited, and the deluge of AI-generated code is making it harder to keep production secure and reliable without slowing you down.
I'm building LiveReview, a blast-radius aware AI code review built for your business-critical systems.
Instead of presenting every diff with equal emphasis, LiveReview scores each change by blast radius — how far its impact reaches through your call graph — so you can focus attention where it actually matters.
Spend code review effort where business risk is highest — not spread evenly across every diff.
⭐ Star it on GitHub:
HexmosTech
/
LiveReview
Blast-Radius Aware AI Code Review for Business-Critical Systems
LiveReview: Blast-Radius Aware AI Code Review for Business-Critical Systems
LiveReview is an AI code reviewer that scores every hunk of a diff by blast radius: how far a change reaches through your call graph, how much persistent state it touches, and how well-tested it is. A 3-line change to a shared auth check can outrank a 300-line UI tweak. Your team's attention goes to the highest-risk code first, not spread evenly across every diff.
blast-radius-demo.mp4
LiveReview's Blast Radius & Review Priority scoring, live in the diff viewer.
| The exact math, not a black box | Visualize blast radius at a glance | Every factor that feeds the score |
|---|---|---|
 |
 |
 |
How does Blast Radius scoring work? (a more technical explanation)
Here's the goal:
- A 3-line fix in a function used by 40 other files, that also writes to a database, should score high.
- A 300-line UI change in one file, fully covered by…
Click below to try LiveReview with your codebase:
Top comments (0)