Case study 01 · Tulasea Inc.
Enterprise Generative AI Platform
Enterprise-grade AI platform architecture combining LLMs, RAG pipelines, vector databases, semantic search, knowledge graphs, conversational AI and workflow orchestration for healthcare and enterprise decision ecosystems.
- Company
- Tulasea Inc.
- Role
- AI Product Architect
- Timeframe
- May 2024 – Present
- Location
- Remote
- Generative AI
- RAG
- AI Agents
- Knowledge Graph
- Enterprise Architecture
01The challenge
Healthcare and enterprise decisions depend on context spread across EHR platforms, operational systems, external APIs and distributed applications. A language model on its own cannot reach that context, reason over relationships in it, or be trusted with it.
The platform had to ground AI in enterprise knowledge, connect to the systems where work happens, and meet the governance, privacy and compliance expectations of a regulated domain, while staying scalable as adoption grows.
02Keerthi's role
As AI Product Architect, Keerthi owns the platform architecture: from AI and data architecture through integration, security and governance, and the standards teams use to build on it.
- Architecting enterprise AI and digital transformation platforms on cloud-native, distributed architecture.
- Defining the AI governance architecture: authentication, authorization, explainability, observability, auditability, data privacy and regulatory compliance.
- Establishing architecture standards, reusable integration patterns, documentation frameworks and design principles for scalable AI adoption.
- Evaluating emerging technologies and defining the architecture roadmap for platform evolution.
- Aligning product, engineering, operations, infrastructure, cybersecurity and executive stakeholders.
03Architecture / approach
The platform treats models, knowledge and workflows as separate, composable layers, so each can evolve without destabilizing the others.
- 01
Retrieval-grounded intelligence
LLMs are grounded through RAG pipelines, vector databases and semantic search, so answers come from enterprise knowledge rather than model memory alone.
- 02
Graph-driven context
JanusGraph models relationships between entities while Qdrant handles vector similarity, together supporting relationship-aware recommendations and contextual reasoning.
- 03
Orchestrated workflows
Workflow orchestration and conversational AI turn model output into AI-assisted workflows, backed by microservices that deliver real-time contextual intelligence and recommendations.
- 04
Secure API orchestration
A secure API orchestration framework integrates enterprise systems, EHR platforms, operational systems and external APIs behind consistent identity and access controls.
- 05
Governance by design
Authentication, authorization, explainability, observability, auditability and data privacy are defined at the architecture level and applied across every layer.
04Technology
AI & Knowledge
- LLMs
- RAG
- Vector Databases
- Qdrant
- JanusGraph
- Semantic Search
- Conversational AI
Services & Data
- Python
- FastAPI
- Microservices
- Kafka
- Databricks
- Event-Driven Architecture
Platform
- Kubernetes
- Docker
- AWS
- Azure
Security
- OAuth2
- OIDC
- REST APIs
- Audit Logging
05Enterprise value
A governed foundation for enterprise AI: one architecture that product and engineering teams can build on, rather than a collection of disconnected pilots.
- Real-time contextual intelligence, recommendations, operational insights and AI-assisted workflows delivered through microservices.
- Relationship-aware recommendations and contextual reasoning through graph-driven intelligence.
- Secure interoperability with EHR platforms, operational systems and external APIs.
- Architecture standards, reusable integration patterns and a technology roadmap for scaling AI adoption.
Outcomes are described qualitatively. Figures appear only where they are verified.
[ADD PROJECT METRIC IF AVAILABLE]06Architecture diagram
Experience
- Conversational AI
- AI-assisted workflows
- Enterprise apps
Orchestration
- Workflow orchestration
- AI agents
- API orchestration
Intelligence
- LLMs
- RAG pipelines
- Semantic search
- Recommendations
Knowledge
- Qdrant · vectors
- JanusGraph · graph
- Databricks
Integration
- FastAPI services
- Kafka events
- EHR platforms
- External APIs
Platform
- Kubernetes
- Docker
- AWS
- Azure
07Key takeaways
- 01
Retrieval quality is an architecture decision. How knowledge is indexed, related and secured matters as much as which model sits on top.
- 02
Vectors find what is similar; graphs explain how things relate. Contextual reasoning benefits from both.
- 03
Governance is cheapest when it is part of the first diagram, not a review at the end.