Update README.md

README.md

@@ -13,23 +13,11 @@ pinned: false
 
 > "Intelligence that speaks our language, not just translates it."
 
-## About
-Dataflare is an engineering-first AI research lab building sovereign intelligence infrastructure for the Middle East. We address the hidden "Arabic Tax" in global AI models by forging cognitively native infrastructure—from silicon to prompt.
+Dataflare is an engineering-first research lab building sovereign AI infrastructure for the Middle East. We address the "Arabic Tax" by forging cognitively native models—from silicon to prompt.
 
-## Core Initiatives
+**Core Initiatives:**
+*   **DF-Arc Tokenizer:** Morphology-aware tokenization reducing Arabic token usage by ~40%.
+*   **Sovereign Datasets:** Curated high-fidelity corpora (Legal, Dialectal, Cultural) replacing web noise.
+*   **Native Alignment:** Models aligned by local domain experts to preserve cultural depth.
 
-### 1. The Tokenization Tax (DF-Arc)
-**The Problem:** Standard models (like `cl100k_base`) fragment Arabic into 2-3x more tokens than English, inflating inference costs and latency.
-**The Solution:** `DF-Arc` is a morphology-aware tokenizer that achieves near 1:1 word-to-token ratios, reducing token usage by ~40% and unlocking higher throughput.
-
-### 2. Sovereign Datasets
-We curate high-fidelity, legally compliant corpora rather than scraping common web noise. Our data pipeline covers:
-*   **Legal Archives** (MSA)
-*   **Dialectal Transcripts** (Egyptian, Gulf, Levantine, North African)
-*   **Cultural Heritage** (Literature, Poetry, History)
-
-### 3. Native Alignment
-Our models are aligned by native domain experts—lawyers, doctors, and poets—to ensure reasoning reflects local reality and values, avoiding the "cultural flattening" of translated English.
-
----
-**© 2026 Dataflare Lab.** Sovereignty is non-negotiable.
+**Contact:** [dataflare.com](https://dataflare.com) | hello@dataflare.com