Ashu Soni

I build production AI features for The Leadership Thread, owning the full AI surface from architecture and implementation through deployment. My work combines Claude and OpenAI with RAG and tool-using agents to automate internal workflows.

I've independently run dev, test, UAT, and production environments on AWS EC2 using FastAPI, systemd, Nginx, and Tailscale, while continuing to own a Django, React, TypeScript, MySQL, and AWS platform. I also reduced recurring inference costs by moving appropriate workloads from paid APIs to self-hosted models on GPU-backed Ubuntu servers using Ollama and vLLM.

Across Vinove and Data Science Wizards, I delivered GenAI features from prototype to production, built LangChain pipelines, RAG chatbots, and ML features for UnifyAI. I work comfortably across model evaluation, backend systems, cloud deployment, and product delivery.

My projects include AirRAG, a private enterprise document intelligence platform serving 500+ documents across 100+ field users, and JAAGI, a local-first voice AI agent with GPU-based speech processing, MCP tools, and a policy-driven security model. I've also built multi-agent healthcare documentation and enterprise voice-and-chat platforms with structured outputs, real-time speech, resilient runtimes, and tenant-aware deployments.