270M text-only EmbeddingGemma-2 fork for Hinglish Hindi RAG. Weights, GGUF, legal adapter, triplets. Apache-2.0. By Eulogik.
AI & ML interests
Edge AI and intelligent super usable systems
Recent Activity
View all activity
Papers
NanoForecast v0.5: Competitive Time Series Forecasting Through Training Pipeline Optimization
Safe Error Correction for Language Models: Frozen-Base Adjustment with Capability Preservation
Cognitive Resonance Network (CRN) adapters achieving 100% exact factual recall via EpisodicMemory tables on frozen backbones.
Next-gen Indic edge LLMs (Bharat-Tiny-LLM), PolyWhisper ASR, and the Brahmi Tokenizer with 33-45% compression for Devanagari.
Federated LLM routing middleware with human-interpretable preference tables. Slashes enterprise API token spend by 40-70%.
Sub-millisecond, CPU-runnable models: pico-type (ArXiv:2608.14658), TinyDoc-VLM (256M), and NanoForecast for Raspberry Pi.
A curated suite of edge-optimized, hyper-efficient, and local-first AI models and interactive spaces. Engineered by Eulogik.
270M text-only EmbeddingGemma-2 fork for Hinglish Hindi RAG. Weights, GGUF, legal adapter, triplets. Apache-2.0. By Eulogik.
Federated LLM routing middleware with human-interpretable preference tables. Slashes enterprise API token spend by 40-70%.
Cognitive Resonance Network (CRN) adapters achieving 100% exact factual recall via EpisodicMemory tables on frozen backbones.
Sub-millisecond, CPU-runnable models: pico-type (ArXiv:2608.14658), TinyDoc-VLM (256M), and NanoForecast for Raspberry Pi.
Next-gen Indic edge LLMs (Bharat-Tiny-LLM), PolyWhisper ASR, and the Brahmi Tokenizer with 33-45% compression for Devanagari.
A curated suite of edge-optimized, hyper-efficient, and local-first AI models and interactive spaces. Engineered by Eulogik.