Print both errors
Univariate and spatial scores both appear. Hiding one to win a procurement cell is not allowed.
Compiling the hour
Compiling the hour
Company
We are not a marketplace for rides. We are not a paint job on Dijkstra. We are the team that treated congestion as a sequence, a graph, a picture, and a policy — then put all four on the same desk.
Places
The graph is here. The 60 days are here. The flyover that lies and the industrial road that does not — here. Kerala is not a backdrop. It is the dataset.
The standard of craft is here. Parchment, not sterile white. Space Grotesk at 300, not a shout. One orange. If it would not hang in a South Park studio, it does not ship.
Why we exist
Every major navigation product will tell you about the jam you can already see through the windshield. That is a weather app for a storm that has started. We wanted the hour before — and a policy that does something with it besides a red polyline.
The work began as a hybrid deep-learning framework: LSTM, GAT, YOLOv8, PPO. It remains that. The company is the decision to treat the framework as infrastructure, not a semester.
Groq taught the industry that speed can look like paper. We took that lesson literally. Bone and parchment for reading. Graphite bands for proof. Ember reserved for the thing you press. IBM Plex Mono for the serial numbers. No drop shadows. No decorative chrome. Variants — parchment and obsidian — instead of a carnival of themes.
We will not hide a worse route behind a prettier chart. We will not pretend a synthetic grid is a city. We will not use vapor pink as a personality. It exists for a rare mark, then it sits down.
The lab still runs Flask, PyTorch, OSMnx, Folium. The public face runs Next.js 16, Tailwind v4, React 19, Three.js. Both directories live under traffic-app because splitting the story from the experiment is how numbers go soft.
Operators, researchers, and engineers in the same review. If a sentence cannot survive all three, it does not go on the site. If a model cannot change a path, it does not go in the loop.
We hire for taste in residuals and taste in type. If you have only one, teach the other here.
Named places from OpenStreetMap. A drive graph you can color by predicted speed. Scenarios for rain and incidents. Routes that can optimize time, not only distance. An influence view that shows which nearby roads are linked. A detect rail for camera density. The capital is the case study; the loop is the product.
Interactive speeds in this deployment are demo forecasts (date + time + scenario), not a live city feed. Geometry and places are real OSM. Detection on uploaded photos is real inference for that image. We label that on purpose so students and operators do not confuse a lab with a municipal CCTV wall.
Principles
Univariate and spatial scores both appear. Hiding one to win a procurement cell is not allowed.
Interactive speeds are scenario forecasts unless a live feed is explicitly connected.
Ember is for actions. The rest of the UI stays parchment and graphite.
Thiruvananthapuram is the case. Portability is earned after the capital graph works.
Detection is density, not policing. Privacy and retention are part of the product conversation.
If a forecast cannot change a path, it is a chart — not infrastructure.
Common questions
No. We build prediction, attention, detection, and routing policy for corridors and labs.
The graph and case study are Thiruvananthapuram. Craft standards borrow from rigorous product design culture — parchment, not noise.
Use the site, then Contact with one failing clock and whether you have cameras or only OSM.