Niyati Patel
Data and AI Engineer. Agents, RAG and BigQuery pipelines on Google Cloud.
Hyderabad, India
patelniyati59@gmail.com
I build AI agents that work on real enterprise data, mostly on Google Cloud. That means Google ADK agents deployed to Vertex AI Agent Engine, MCP tool services on Cloud Run, retrieval with Vertex AI Search, and BigQuery underneath. I have worked across healthcare, health insurance, government, higher education and manufacturing. Some things I have built: A multi agent data orchestrator that routes natural language questions to BigQuery, Spanner, Cloud SQL and document search, with schema discovery, source citations and a confirmation step before any write. A Databricks agent that discovers Unity Catalog schemas and runs read only SQL under each user’s own OAuth token, so existing data permissions still apply. A sourcing agent for a county government that refines draft bid specifications against the approved template, then finds suppliers from past bids and public web research. A records agent that reaches OneDrive and on premise file shares over VPN, returning only what the user already has access to. I like the full path: discovery with the client, writing the architecture, building it, and getting it into production.
Software Engineer, AI and Data
Zion Cloud Solutions
- Build and deploy AI agents on enterprise data for clients in healthcare, insurance, government, higher education and manufacturing, from discovery through to production.
- Design multi agent systems in Google ADK, deployed to Vertex AI Agent Engine and delivered inside Gemini Enterprise.
- Connect agents to BigQuery, Databricks, Spanner, Cloud SQL, SharePoint, OneDrive and on premise file shares, keeping each user’s existing permissions intact.
- Build MCP tool services and data pipelines on Cloud Run, with Terraform for infrastructure.
- Run client discovery sessions and write the architecture documents that scope each build.
Software Developer
Premier IT Solutions
- Built web applications in Python, Flask and PostgreSQL processing over 100 GB of data monthly.
- Implemented REST APIs and CI CD pipelines with Docker, Kubernetes and Jenkins, improving deployment efficiency by 50 percent.
- Added communication and synchronization mechanisms that reduced errors by 10 percent.
Software Engineer
Accenture
- Automated over 100 scripts at 90 percent accuracy and contributed to backend development in Java and Python.
- Led end to end QA, documenting over 100 defects.
- Trained and mentored more than 10 team members.
Multi agent data orchestrator
A root agent that routes natural language questions to five specialist agents across BigQuery, Spanner, Cloud SQL, document search and conversational analytics. It discovers table schemas at run time, cites the engine and table behind every answer, and asks for confirmation before any write. Delivered inside Gemini Enterprise as an agent to agent service on Cloud Run, with tables, charts, forms and downloadable PDF and Word reports. Built for a healthcare provider.
Databricks lakehouse agent
An agent that answers questions over a Databricks lakehouse using read only SQL. Each user authorizes once and every query runs under their own token, so Unity Catalog grants decide what they can see. It walks catalogs, schemas and tables to discover structure before writing a query, refuses any write statement, and also converts between SQL and GraphQL. Built for a health insurance client.
Sourcing and supplier agent
Refines a draft bid specification against a county’s approved template and flags gaps, then after the user approves, finds candidate suppliers from past bids, the vendor list and public web research, with a source behind every recommendation. Exports the refined spec to Word and the supplier list to Excel. Deployed two ways, to Vertex AI Agent Engine and to Cloud Run with interactive result cards in Gemini Enterprise. Tech used: Google ADK, Gemini 2.5 Pro, Vertex AI Agent Engine, Gemini
Budget and invoice agents
Two agents over county finance and ERP data. One tracks budget spend down against itemized fiscal year budget lines, the other reviews invoices. Both map transactions to the correct budget line and say plainly when no matching line exists rather than guessing
Master of Science, Computer Science
SUNY Polytechnic Institute, Utica, NY · 3.72/4
Bachelor of Engineering, Electronics and Communication
L.D. College of Engineering, Gujarat, India · 8.52/10
Google Cloud Professional Machine Learning Engineer
Google Cloud Professional Cloud Architect
Attendee · Oct 24, 2026