I'm Arun Baruah, an Enterprise AI Consultant and GenAI Automation Engineer helping startups and businesses build domain-specific AI solutions — fine-tuning open-source LLMs (Llama, Qwen, Gemma, Mistral, DeepSeek), developing secure on-premises RAG systems, designing multi-agent AI architectures, and automating complex business workflows without exposing sensitive data to public AI services.
Scalable AI applications designed to drive real-world business value.
LoRA, QLoRA, PEFT, and SFT training pipelines using Unsloth and Hugging Face. Tailoring models like Llama, Qwen, Gemma, Mistral, and DeepSeek for specific domain expertise.
Multi-agent AI systems, enterprise knowledge assistants, and document intelligence built with LangChain, LangGraph, LlamaIndex, and vector databases like FAISS, Qdrant, and Milvus.
End-to-end intelligent workflow orchestration using n8n, Make, and Zapier — CRM automation, lead management, and business process automation to eliminate repetitive manual work.
PDF parsing, OCR pipelines, invoice automation, and structured data extraction, tailored for BPO/KPO and enterprise document-heavy workflows.
Gmail automation, AI email assistants, email classification, auto-reply systems, and inbox management to keep communication moving without manual triage.
Deploying AI applications on Azure, AWS, or GCP with Docker and Kubernetes. Model serving via vLLM and Hugging Face, plus ongoing monitoring and scaling.
Continuously validated skills through elite, hands-on programs.
Advanced agent workflows, persistence layers, state management, and human-in-the-loop systems with LangGraph.
Quantization mechanics, dataset curation pipelines, prompt tuning, and hyperparameter tuning optimizations for small to mid-sized models.
Designing, implementing, and deploying enterprise-grade Cognitive Services, custom models, and Azure OpenAI infrastructure securely.
Specialized AI engineering engagements for building secure, production-ready enterprise AI systems.
A production-ready retrieval system built around your internal PDFs, documents, and data sources — with secure data ingestion, access-scoped retrieval, vector search, and an architecture designed to integrate with your existing enterprise workflows.
Purpose-built multi-agent systems designed around your business workflows — enabling specialized agents to plan, reason, delegate, use tools, and execute complex tasks with stateful orchestration and human oversight where required.
Dataset preparation, synthetics generation, cleaning, and model evaluation pipeline. Delivery includes a quantized model ready for production serving.
Have an automation roadblock or need a specialized private LLM architecture? Send over your context below.