# nadlens

> nadlens is a visual geolocation service by nadcode (powered by the GeoViT engine) that tells you where a street-level photo was taken and shows the evidence behind the answer. It is in development: the training-free pipeline is being validated and a public demo is coming soon.

## How it works

- Read the scene: street signs and names are read and matched against OpenStreetMap.
- Reason about the clues: a vision language model proposes a country, city and area with its reasons.
- Match against a region pack: the photo is compared with geotagged Mapillary street-level images to narrow the answer to a neighbourhood.
- Decide: a final step returns a location, a confidence circle and the evidence list.
- Region packs (reference imagery, embeddings, boundaries, language hints) replace model retraining. Istanbul is the first pack; Türkiye and global coverage follow.

## Modes

- Find: where was this photo taken?
- Verify: was this photo taken at the claimed address?

## Responsible use

- Built to locate places, not people; photos whose main subject is a person are declined.
- Answers are neighbourhood level with a confidence circle.
- Uploaded photos are not stored.
- Reference imagery: Mapillary contributors, CC-BY-SA 4.0.

## Roadmap

0. Validation (current), 1. Public demo, 2. API beta, 3. Türkiye, 4. Global.

## History

GeoViT v1 was a trained Trinity architecture for Istanbul (75% single model, 81-83% with the four-layer cascade). It is documented as a research archive.

## Pages

- [Home](https://lens.nadcode.com/)
- [GeoViT v1 research archive](https://lens.nadcode.com/under-the-hood/)
- [Privacy notice](https://lens.nadcode.com/privacy/)
- Team: nadcode (https://nadcode.com): Elber Dalfidan, Mert Can Batur