Architecture decisions, edge AI model training, and systems engineering by Achyut Srivastava (14) & Shubham Dangi (15) at NexInova.
How we pre-trained and fine-tuned a 57.2M parameter edge transformer with 22 layers, MQA (7Q/1KV), SwiGLU, and built-in Program-of-Thought (PoT) tool execution on dual Tesla T4 GPUs.
Deploying hybrid Edge-TTS and Cloudflare Worker routing to deliver 1,000+ words in 11 seconds with zero SaaS subscription markups and transparent 20 LC/hr metering.
High-DPI coordinate mapping, visual grounding, and multi-monitor screen perception for the LuxurAI Cowork agent running on an in-house dual-GPU cluster.
Eliminating credential databases entirely: how ephemeral signed tokens and HMAC validation protect user sessions with zero stored passwords.
How LuxurAI Coder catches bracket mismatches, name drifts, and syntax errors in real-time sandbox loops before touching customer codebases.