Malav Naik: From Data Engineering to Knowledge Graph-Driven AI Systems

From Gujarat, India to Dublin, Ireland

Malav Naik's career began in data. Over more than five years at S&P Global Market Intelligence, he worked as a data analyst and ETL developer, building SQL data pipelines, migrating databases to the cloud, and ensuring that business intelligence reports rested on a reliable data foundation. That experience drove home a key lesson: AI reliability is always only as good as the data engineering underneath it.

In September 2024, he moved to Dublin to pursue a master's degree in data analytics at the National College of Ireland, graduating in 2025 with a 2:1. His thesis focused on how knowledge graphs and prompt engineering can support AI-driven longitudinal analysis—in short, giving language models structured factual grounding rather than letting them guess. This research direction became the core philosophy behind the AI systems he went on to build.

The combination of knowledge graphs and prompt engineering aims to solve a fundamental limitation of large language models: model knowledge is frozen in training data and cannot guarantee factual accuracy or timeliness. Knowledge graphs store entities and their relationships in a graph structure, providing verifiable structured facts; prompt engineering, meanwhile, guides models through designed inputs to reason over those facts. In longitudinal analysis scenarios, this means tracking how the same entity (a patient, a company) changes over time and enabling the model to answer based on the timeline in the graph. This approach stops language models from "guessing" and instead has them generate answers as if consulting a structured archive, significantly improving the interpretability and trustworthiness of outputs.

Current Work: Landing AI Automation in Healthcare

Today, Malav Naik works in Dublin as a BIS analyst and application analyst/developer at Trinity Care Nursing Homes. The AI-driven automation and data analytics solutions he designs support the operations of 13 nursing homes. His core work includes consolidating operational reporting, integrating business systems, and testing an AI integration with Oracle NetSuite—where authorized users can ask financial questions in natural language, and the system safeguards security through role-based access control and human approval for sensitive operations.

Through his work, he has repeatedly validated a key principle: most of the effort is actually data validation, audit logging, exception handling, and determining where human intervention must remain. This "ask before acting" approach to AI automation design offers a valuable reference for any team deploying AI agents in enterprise environments.

Role-based access control (RBAC) is a method of managing user access to system resources by defining roles and their permissions. In the Oracle NetSuite AI integration at Trinity Care Nursing Homes, RBAC ensures that employees in different positions (such as finance, nursing supervisors, and administration) can only access the financial data and operations within their scope of responsibility. Combined with "human approval for sensitive operations," the system can automate high-frequency, low-risk tasks while triggering human review for high-risk actions such as payments or data modifications. This design balances efficiency and compliance—especially in heavily regulated industries like healthcare, where it prevents AI agents from causing irreversible business risks through misjudgment or overreach.

SkillBot: A Deep Integration of RAG and Knowledge Graphs

During his master's studies, Malav Naik built SkillBot—a skill recommendation chatbot that combines a fine-tuned GPT-3.5 model, retrieval-augmented generation (RAG) technology, and a knowledge graph containing roughly 2.3 million RDF triples.

By anchoring answers to the knowledge graph, the system achieved 97% factual accuracy in reported evaluations and significantly reduced hallucinations. The technical approach here is worth noting: rather than relying solely on a larger model, it uses structured knowledge to constrain the language model's output, finding a practical balance between accuracy and trustworthiness.

RDF triples are the basic data units of the Resource Description Framework, representing facts in a "subject-predicate-object" form—for example, (aspirin, treats, headache). 2.3 million triples means the knowledge graph stores roughly 2.3 million such structured facts in the medical or skills domain. RDF's advantage lies in its high degree of standardization and inferability: based on triples, ontologies can be built and logical reasoning performed to infer implicit relationships. In SkillBot, RDF triples serve not only as a retrieval corpus but are also invoked by the RAG pipeline through SPARQL queries or vectorized indexing, providing the language model with precise, traceable factual grounding—the structural foundation for achieving 97% factual accuracy.

VEDASRA: Creative Expression Beyond Technology

Beyond work and academic research, Malav Naik is also building an anime-style creative project in Dublin called VEDASRA. Though details released so far are limited, it reflects another side of him as a technology builder—combining systems thinking with creative expression.

What He Plans to Share

Malav Naik says he will share lessons from real work every week, including:

• How SkillBot combines RAG with knowledge graphs
• Practical insights on prompt engineering from his thesis
• Lessons from five years of enterprise-grade ETL work
• How to design "ask before acting" AI automation systems
• Behind-the-scenes stories from the VEDASRA project

For practitioners working in AI agents, RAG, or data engineering—especially in Ireland's tech community—Malav Naik's journey offers a practical perspective worth following: beyond model capabilities, data engineering and system design are the true foundation for putting AI into production.

Malav Naik · Dublin · Creator of VEDASRA Portfolio: malavnaik99.github.io · GitHub: malavnaik99