About
I'm Raghunath Mahakud, a GenAI Engineer with 10+ years of IT experience — 7 years as a Backend Java engineer and 4 years deep in Generative AI, LLMs, and NLP. Currently a Senior Software Engineer at Pegasystems, building the AI systems that power enterprise decisioning at scale.
My focus: LLM fine-tuning, RAG architectures, Agentic AI, and multi-agent systems. I've designed autopilot assistants, production NLP platforms, and IoT microservices for enterprise clients including SoftBank, Mercedes-Benz, and the Government of Rwanda.
Currently pursuing M.Tech in AI & ML at BITS Pilani (WILP), deepening expertise in the theory behind the systems I build every day. Active on Hugging Face — fine-tuning and publishing open-source models, and a member of PyTorch Day India HF.
Selected Work
Enterprise-grade systems across GenAI, NLP, IoT, and automotive — built for global clients.
LLM-driven task orchestration with fine-tuned models, plugin architecture, and multi-step planner agent for enterprise decisioning.
Production NLP for topic detection and entity extraction with multilingual support serving enterprise clients globally.
Geo-redundant platform handling OTA firmware upgrades, diagnostics, and monitoring at scale.
Microservices migration with event-driven Kafka architecture serving all Mercedes-Benz car testing globally.
Multi-service e-Government portal with cross-department integration for national service delivery.
NLP-powered product discovery engine combining domain knowledge graphs with Elasticsearch.
Open Source
Published models, live demos, and learning repositories across the AI/ML open-source ecosystem.
Fine-tuned Qwen3 model for English to Odia translation — 420+ downloads. Published with a companion Spaces demo.
View on Hugging Face →Interactive Gradio Space — translate English sentences into Odia instantly. Powered by the fine-tuned Qwen3 model.
Try it Live →Implement a ChatGPT-like LLM in PyTorch from scratch — deep understanding of transformer internals.
View on GitHub →DeepSeek implemented from scratch to understand MoE, attention mechanisms, and modern LLM architecture internals.
View on GitHub →Implementation of 17+ agentic architecture patterns — from ReAct and tool-use to multi-agent orchestration systems.
View on GitHub →Advanced fine-tuning techniques — LoRA, QLoRA, full fine-tuning, and optimization strategies for any LLM.
View on GitHub →MCP (Model Context Protocol) server for Apache Kafka — enabling AI agents to interact with Kafka infrastructure.
View on GitHub →Collection of different RAG techniques — naive RAG, advanced retrieval, re-ranking, hybrid search, and evaluation.
View on GitHub →Career Path
Expertise
Deep stack spanning Generative AI, distributed systems, and production engineering — 10+ years across 4 enterprise domains.
Credentials
10 verified credentials across GenAI, NLP, RAG, RL, and enterprise architecture.
Get In Touch
Open to GenAI consulting, LLM architecture discussions, and enterprise AI collaborations.
Whether it's about Agentic AI architectures, LLM fine-tuning, or scaling production NLP systems — I'd love to connect. Member of PyTorch Day India HF community.