Distributed AI Ranked Network (DARN)

Inspiration

Was inspired by the Shodan Search Engine which allows users to search for various types of servers connected to the Internet using a variety of filters.

What it does

DARN tracks and ranks AI nodes running across the internet and answers the question "Which AI endpoints are alive, fast, and actually usable right now?".

Core functions

1. Network health monitoring

Tells users whether the network is reliable at that moment by classifying them as healthy (responding) or unhealthy (failing or unreachable) and calculates a success rate.

2. Latency benchmarking

Measures response time for each responsive node
Shows average latency and fastest node This is critical for routing and load balancing

3. Model discovery

Scans verified hosts to detect which models are actually deployed and how frequently each model appears

4. Node-level inspection

For each IP, the site tracks

Last check timestamp
Status (OK/ FAIL)
Latency
Models host (or "no models")

5. Failure tracking

Shows list of failing nodes
"View details" feature to show logs or diagnostics
Helps operator identify offline machines, empty hosts and broken deployments

6. Geographical visualisation

Map view (using Leaflet and OpenStreetMap)
Shows where AI compute is physically distributed

7. Inventory snapshot

Node -> Models -> Latency
Highlights ghost nodes ("no models") and valuable nodes (fast + multiple models)

Our platform is useful for people running decentralised AI networks and developers routing inference across multiple providers.

How we built it

Used Python, React, and Tailwind

Challenges we ran into

We initially chose the travel track where our idea was to gamify a trip to maximise cultural exposure to a city/ country, where users are given a list of tasks, and when each task is completed, they will gain points and a leaderboard is formed. 6 hours into the hackathon, we realised that the idea was flawed as if users are travelling with a group of friends, they are likely to do the same things together, therefore the leaderboard doesn't have much meaning.

Therefore, we switched to this idea relatively late into the hackathon, which gave us a tighter time constraint of what can be done.

Accomplishments that we're proud of

Successfully implementing the core features, smoothly transitioning to another idea that we collectively thought was better.

What we learned

We have learnt how to recognise a flawed solution and quickly transition to another track and build another application that makes more sense conceptually.