Agentic AI & Orchestration
Tool-aware LLM workflows, human-in-the-loop design, retrieval-grounded multi-step reasoning and orchestration patterns.
I'm Harshavardhini Selvaraajan — an AI engineer turning complex ideas into production-ready GenAI products, intelligent workflows, cloud platforms and digital twins.
I work across the layers that make AI useful — models and retrieval, backend systems, cloud infrastructure, and the experience people actually use.
Tool-aware LLM workflows, human-in-the-loop design, retrieval-grounded multi-step reasoning and orchestration patterns.
Grounded assistants that connect private documents and domain context to reliable search, retrieval and answer generation.
Evaluation, regression testing, prompt/model comparison, deployment quality gates and monitoring for maintainable AI.
Usable, data-backed AI products with APIs, frontend interfaces, business workflow integration and scalable persistence.
AWS-based solution architecture covering event-driven processing, model services and resilient deployment.
Domain hierarchies, visual component selection and contextual data interfaces connecting digital models to real workflows.
Real engineering work across enterprise AI, healthcare and maritime — plus collaborative digital-twin prototypes.
Select any project to explore the architecture, engineering challenges, contributions and outcomes.
Digital twins connect structured domain knowledge, spatial representation and operational context. These collaborative concepts explore what data-connected interfaces can make possible.
01 / PATIENT DIGITAL TWINAn interactive anatomy experience where selected organs, semantic structure and visit history come together in one interface.
A multi-floor spatial interface connecting reception, pharmacy, diagnostics and other services to area-level statuses and simulated operational insights.
02 / HOSPITAL OPERATIONS TWINDigital twin examples are collaborative demonstrations. Hospital interface data is simulated; they are not presented as validated clinical systems.
The hardest part of AI isn't getting a promising demo. It's building the data flows, evaluation, APIs, interfaces and operational practices that let people rely on it.
Let's build somethingStart with the business workflow, the users and the data.
Design the retrieval, models, storage, APIs and controls.
Deliver a usable full-stack system, not an isolated notebook.
Test quality, operate reliably and iterate after launch.
A cross-domain career combining applied machine learning, enterprise GenAI, cloud engineering and hands-on product delivery.
Enterprise LLMOps/MLOps, GenAI evaluation, AI-powered CRM and architecture.
Clinical coding intelligence and semantic extraction workflows.
Maritime reporting AI assistant and production-oriented domain AI.
Cargo planning optimization and decision support.
Explainable AI research for digital pathology and whole-slide images.
From domain understanding to working software, every layer matters.
I'm Harsha, an AI engineer with over five years of experience across enterprise technology, healthcare and maritime applications. I enjoy translating ambiguous problems into reliable systems that teams can actually use.
My work lives at the intersection of machine learning, generative AI, software engineering and cloud architecture. I like to think beyond the model: How will data arrive? How do we evaluate quality? What happens when an API fails? And how do we make the experience intuitive?
I'm especially interested in the next generation of AI-native applications, agentic workflows and digital twins — with an emphasis on useful, trustworthy outcomes.
Open to AI engineering opportunities in the UAE and India, consulting engagements and collaborations on ambitious AI products.
harshaselvarajan@gmail.com