Add Qandy AI organization card

README.md

@@ -1,10 +1,37 @@
 ---
 title: README
-emoji: 🐢
-colorFrom: purple
+emoji: 🍬
+colorFrom: pink
 colorTo: indigo
 sdk: static
 pinned: false
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Qandy AI
+
+**Explainable language models for crypto trading desks.**
+
+Qandy AI builds compact reasoning models that read a structured market snapshot and return one clear, checkable decision: go long, go short, or stand aside. Each decision comes with a stop, a target, and the levels behind it. Every call shows its reasoning, and that reasoning cites only the signals it was given.
+
+## Starburst
+
+Starburst is our model family for crypto perpetual futures. Starburst models are trained in two stages:
+
+1. **Supervised fine-tuning** on real desk decisions, which teaches the reasoning style and the decision format.
+2. **KTO preference tuning** on real trade outcomes, which pushes the model away from setups that lost money and, in later versions, away from skipping setups that would have won.
+
+| Model | Size | Base | Release |
+|---|---|---|---|
+| [Qandy 1.0 27B](https://huggingface.co/QandyAI/qandy-1.0-27B) | 27B | Qwopus3.6-27B-v2 (Qwen3.6) | Recipe and evaluation release |
+
+## How we work
+
+- **Grounded.** Models cite only what is in the input, so every claim can be checked.
+- **Outcome-trained.** Preference tuning uses real losing trades, not only winners.
+- **Honest evaluation.** We report results against live desk outcomes, including when a model does not beat its predecessor.
+
+## Links
+
+- Website: [qandy.ai](https://qandy.ai)
+
+*Qandy AI models are research tools, not financial advice. Do not connect them directly to order execution.*