Databases for Agentic AI
Planning, execution, metadata, state, and provenance matter as much in agents as they do in database systems.
Subhasis Dasgupta, Ph.D.
Building the Data and Systems Foundations for Scientific AI
Subhasis Dasgupta is a database-systems researcher and scientific-platform architect at the San Diego Supercomputer Center and UC San Diego. He designs trusted infrastructure that connects heterogeneous data, high-performance computing, knowledge systems, scientific workflows, and AI agents. Earlier in his career, he was the first employee of the U.S. cloud startup Kaavo, where he helped build its multi-cloud management platform and establish its India engineering operation.
Research question: How can intelligent systems plan and operate reliably across databases, APIs, models, computing resources, and scientific instruments while preserving provenance, performance, security, and human oversight?
Trusted across institutions and systems
- UC San Diego
- San Diego Supercomputer Center
- Kaavo
- Database Systems
- Scientific Platforms
- AI-Enabled Discovery
Signature framework
The Scientific Intelligence Stack
Reliable scientific AI starts with query planning, provenance, execution control, and infrastructure that can span data, models, workflows, and instruments.
Scientific AI Infrastructure
Heterogeneous data, HPC, workflows, and knowledge systems need to cooperate without hiding system behavior.
Architecture and Technical Leadership
Complex systems succeed when the technical story is legible to researchers, operators, and collaborators.
Flagship work
Representative projects and contributions
Project
AWESOME Polystore
Heterogeneous database integration and query processing for relational, graph, text, and analytical systems.
Project
National Data Platform Search
Federated scientific-data discovery with metadata, ontology-driven search, and national-scale APIs.
Project
TemPredict
Wearable and multimodal data infrastructure for predictive-health research and clinical support.
San Diego Supercomputer Center / UC San Diego
AWESOME Polystore
A research and platform effort focused on making heterogeneous databases work as one system for scientific and operational workloads.
- Enabled published work on polystore ingestion and query planning.
- Connected to patentable ingestion and query-processing methods.
San Diego Supercomputer Center / National Data Platform
National Data Platform Search
A search and discovery layer for scientific data platforms that emphasizes metadata, semantic access, and community-scale usability.
- Connected distributed scientific catalogs to a reusable discovery layer.
- Aligned with the National Data Platform's AI-ready data and workspace model.
UCSF Osher Center for Integrative Health / UC San Diego
TemPredict
A system focused on ingesting and organizing multimodal health data for analysis, collaboration, and research workflows.
- Enabled a published conference paper and a large-scale wearable-sensing study.
- Connected platform work to symptom-prediction and monitoring workflows.
Research agenda
The questions that keep repeating
How should agents plan across databases, APIs, models, and scientific tools?
Can query optimizers reason about subjective and open-world predicates?
How should scientific digital twins be registered, discovered, executed, evaluated, and archived?
Can database instrumentation generate actionable system recommendations?
Research focus
The work centers on scientific data systems that preserve meaning, manage execution cost, and let both humans and agents make reliable decisions.
- Query planning and optimization across heterogeneous systems
- Provenance, governance, and reproducibility in scientific platforms
- Distributed execution, HPC, and workflow orchestration
- Knowledge graphs, retrieval, and AI-assisted discovery
Publications and patents
Selected research record
journal
Sex Differences in the Variability of Physical Activity Measurements Across Multiple Timescales Recorded by a Wearable Device: Observational Retrospective Cohort Study
journal
Biometrics of complete human pregnancy recorded by wearable devices
journal
General feature selection technique supporting sex-debiasing in chronic illness algorithms validated using wearable device data
journal
Utilizing Wearable Device Data for Syndromic Surveillance: A Fever Detection Approach
conference
P2KG: Declarative Construction and Quality Evaluation of Knowledge Graphs from Polystores
Declarative construction and quality evaluation of knowledge graphs from polystore data.
journal
Detection of COVID-19 using multimodal data from a wearable device: results from the first TemPredict Study
Patent
Data Ingestion into a Polystore
Patent
Query Processing in a Polystore
Blogs
Current essays and notes
Blogs
What I Learned as the First Employee of a U.S. Cloud Startup
Building before requirements stabilize, and how platform decisions shape long-term engineering.
Blogs
Why Scientific AI Needs Query Planning, Not Just Retrieval
Scientific questions require multiple tools, validation steps, and provenance-aware execution.
Blogs
What Database Researchers Can Contribute to Agentic AI
Why planning, provenance, optimization, and data semantics matter for reliable agent systems.
Speaking and workshops
Talks, panels, and collaboration
The speaking page will expand as verified talks, panels, and workshops are added.
Research collaborationAcademic partnershipStartup technology discussionInvited talk or panelTechnical workshopOpen-source collaboration
Leadership story
From first employee to platform builder
Subhasis Dasgupta, Ph.D., is a database-systems researcher and scientific-platform architect at the San Diego Supercomputer Center and UC San Diego. His work spans query processing, heterogeneous databases, distributed systems, scientific workflows, HPC, knowledge graphs, and AI-enabled data discovery. Earlier in his career, he was the first employee of the U.S. cloud startup Kaavo, where he helped build its multi-cloud management platform and establish its India engineering operation.
What I bring
- Research depth in database systems and scientific data infrastructure
- Platform experience across HPC, cloud, and workflow environments
- Technical leadership that connects architecture to execution
- Collaboration with scientists, engineers, and institutional teams
Connect
Let’s build the systems layer for scientific AI.
I welcome conversations with researchers, founders, engineers, and institutions working on databases, scientific data platforms, distributed systems, and AI-enabled infrastructure.