Sourabh Singh

I've built end-to-end AI systems spanning LLM fine-tuning, Retrieval-Augmented Generation pipelines, autonomous agents, and distributed GPU orchestration. My recent work integrates GPU capacity across Akash Network, Cudo Compute, HostKey, FluxEdge, and Octa Compute for cost-aware, dynamic workloads.

I fine-tune Llama, Mistral, Falcon, and Phi models with LoRA and QLoRA, build agentic workflows with LangChain, LangGraph, and LlamaIndex, and operate production inference using vLLM, TGI, and ONNX Runtime. I've also created LLM evaluation frameworks using BLEU, ROUGE, BERTScore, perplexity, LangSmith tracing, and human-preference benchmarks.

Previously, I built an Amazon Merch intelligence platform that scraped 100k+ products, forecast sales from BSR data, and moved from the ChatGPT API to self-hosted open-source inference, reducing per-request costs by roughly 80%.

I've also delivered travel and hospitality analytics platforms, Django scraping and payments applications, KPI dashboards, data warehouses, and Stable Diffusion/AnimateDiff generation systems. I'm open to remote full-time roles, long-term freelance contracts, and AI consulting engagements.