Open to opportunities

Puru Koli

@purukoli

Message

I build production Generative AI, RAG, and real-time voice systems for enterprise customer experiences.

India

Message

What I'm looking for

I'm looking to contribute to frontier Generative AI and Physical AI challenges at a hyperscale organization, building scalable LLM, multimodal, and on-device AI systems.

I've built production AI systems for Akasa Air, Ducati, and Bose at NeoSoft Technologies, including an email intent classifier, a RAG-powered WhatsApp bot, and a locally hosted product-support voice bot.

I build end-to-end LLM and voice pipelines, from RAG retrieval and tool calling to STT, TTS, VAD, quantized local models, and scalable vLLM inference. My offline voice bot delivered context-aware multi-turn conversations with sub-second latency and no external API dependencies.

Previously at Qubrid AI and Koderbox Technologies, I deployed RAG chatbots on Google Cloud Run, researched fine-tuning and quantization, and fine-tuned mT5 for Hindi grammar correction to 92.34% precision. I'm motivated by frontier Generative AI and Physical AI challenges at hyperscale organizations.

Experience

Work history, roles, and key accomplishments

Current

AI Engineer

Current

NeoSoft Technologies Pvt. Ltd.

Mar 2025 - Present (1 year 6 months)

Built production-grade AI systems including an email intent classifier for Akasa Air, a Ducati WhatsApp bot, and a local voice bot for Bose. Implemented scalable local LLM inference with vLLM and developed a real-time noise cancellation application.

LLM RAG Multi Agent Systems VLLM Redis Python Docker Speech to Text Text To Speech Voice Activity Detection

AI Engineer

Qubrid AI

Feb 2024 - Feb 2025 (1 year)

Led POC research on no-code LLM and Stable Diffusion fine-tuning workflows. Architected a conversational AI system and deployed production chatbots on Google Cloud Run with Milvus vector database.

LLM RAG Stable Diffusion Google Cloud Run Docker Milvus TensorRT Python TTS VAD

Artificial Intelligence Engineer

Koderbox Technologies Pvt. Ltd.

Apr 2023 - Feb 2024 (10 months)

Researched and developed AI models end-to-end, including fine-tuning Google's mT5 for Hindi grammar correction and building a Manga-style text-to-image pipeline with Stable Diffusion.

Fine Tuning Stable Diffusion Python NLP Deep Learning