Thanush P K
Building the agentic layer between AI and real work.
~$
Four-plus years across software engineering — now leading GenAI solution delivery within the NTT DATA Group, on a foundation of Java & Spring backend work.
Profile
An AI engineer from India, based in Nagoya, Japan, working at the seam between applied LLMs and production backend systems. After 3+ years building Java/Spring services for enterprise e-commerce, I moved into GenAI — and now lead a small team delivering custom AI agent solutions for large enterprise clients, including in automotive and manufacturing.
My focus is the unglamorous part that makes AI actually useful: orchestrating agent workflows, grounding them in RAG over real knowledge bases, hardening tool-calling through LoRA fine-tuning, and integrating the gateways and protocols (MCP) that hold the whole system together. I work comfortably inside a Japanese enterprise environment in both English and Japanese.
Experience
AI Engineer — Team Lead
- Lead a small AI engineering team delivering custom AI agent solutions for large enterprise clients, including in the automotive / manufacturing sector.
- Design and orchestrate end-to-end LLM-powered agent workflows on Dify, tailored to client requirements, integrating RAG pipelines over enterprise knowledge bases.
- Evaluate and integrate emerging GenAI tooling — AI gateways, self-hosted / open-source LLMs (Docker), and agent protocols (MCP) — on a local GPU environment.
- Drive LoRA fine-tuning to improve tool / function-calling reliability, and mentor the team on AI development and testing practices.
Software Developer
- Built and maintained Java/Spring backend services for an enterprise e-commerce platform serving a global travel client.
- Migrated legacy CMS components to a Micro Frontend (MFE) / Backend-for-Frontend architecture using GraphQL, TypeScript, and a headless CMS.
- Delivered a proof-of-concept migrating merchandising rules into a decoupled service, integrating third-party APIs and presenting to stakeholders.
Stack
AI / GenAI
Backend
Frontend
Tools, Cloud & Practices
Selected Work
Guarded Agent: Per-Call Auth & Learned Constraints over MCP
An AI agent querying databases via MCP with per-tool policy verification and memory-based constraint learning using pgvector.
Llama 3.2 1B Fine-Tuning (QLoRA)
Fine-tuned Llama 3.2 1B with QLoRA — parameter-efficient instruction tuning on consumer-grade hardware.
RAG vs OKF: A Retrieval Benchmark
A retrieval benchmark comparing standard RAG against an Open Knowledge Formatter preprocessing step.
Employee Management System
A full-stack web application serving a RESTful API with an Angular frontend and responsive design — built end to end on a Spring Boot backend.
Credentials
Certifications
Languages
in a Japanese enterprise setting
Education
77% · 2020–2022
60% · 2017–2020