On-Device AI & Systems Engineer · Apple Silicon Inference · Founder, The Smart Language

I build production apps and the inference runtimes underneath them — local LLMs, VLMs, and speech models running entirely on-device on Apple Silicon (Neural Engine + Metal). 20K+ commits across 50+ repos in three years. 2× international AI hackathon winner.

🧠 On-device inference — custom hybrid ANE/GPU runtime; LLMs/VLMs/STT/embeddings that run 100% on the Neural Engine or Metal, at 1–2 W, no cloud in the hot path.

⚡ Model engineering — 8-bit & ternary (BitNet b1.58) quantization, LoRA adapters, CoreML conversion, per-op ANE residency tuning. e.g. LFM2-VL-450M running fully on the ANE.

🎙️ Voice AI — real-time STT → LLM → TTS pipelines; WebRTC (Pion/Go), LiveKit, realtime APIs, Twilio telephony. Built a WhatsApp Voice AI bridge in pure Go.

🛠️ Stack — Rust · Swift · Metal · CoreML · MLX · Python · Next.js · FastAPI · Supabase

• The Smart Language — AI platform for language institutes
• Haku — a local-only MCP server that cuts AI agents' token usage ~85% with on-device filtering
• SuperTutora — a hyper-personal language tutor

📍 Hyderabad, India · shahir.thesmartlanguage.com · @TslShahir