Agentic Sciences Team

Fujitsu Research of India

Research Philosophy

We take a problem first approach to ML research, especially the sub-field of "Agentic" AI. That is, we aspire to build systems which interact with the environment (human, software, nature etc) and get measurably better with the feedback to solve a given problem.

Our Research Contributions

Research Contributions

Our Research Contributions

Our team focuses on developing next-generation agentic AI systems that can reason, plan, and execute complex tasks autonomously. We work at the intersection of large language models, multi-agent systems, and behavioral evaluation frameworks.

Our research spans novel evaluation methodologies for compound AI systems, contextual routing mechanisms for optimized model selection, and frameworks that assess AI behaviors rather than static benchmarks.

Our Team

Chaitanya Devaguptapu

Chaitanya Devaguptapu

Lead Researcher

Suraj Nagaje

Suraj Nagaje

Researcher-II

Harsh Vishwakarma

Harsh Vishwakarma

Researcher-II

Vartika Sengar

Vartika Sengar

Senior Researcher

Shrey Satapara

Shrey Satapara

Researcher - II

Parth Thakkar

Parth Thakkar

Researcher - II

Pranoy Panda

Pranoy Panda

Researcher-II

Ankush Agarwal

Ankush Agarwal

Researcher-II

K.N Ajay Shastry

K.N Ajay Shastry

Researcher - II

Featured Publications

Adaptive LLM Routing Under Budget Constraints

Optimizing model selection through contextual bandit algorithms for efficient compound AI systems.

Click to read more →
EMNLP 2025

Can LLMs Help You at Work? A Sandbox for Evaluating LLM Agents in Enterprise Environments

A comprehensive evaluation framework and sandbox environment for testing LLM agents in real enterprise scenarios.

Click to read more →
EMNLP 2025

Evaluating Compound AI Systems through Behaviors, Not Benchmarks

A novel framework for behavioral evaluation of AI systems beyond traditional benchmark metrics.

Click to read more →
EMNLP 2025

Finding Needles in Images: Can Multi-modal LLMs Locate Fine Details?

Investigating the capability of multi-modal language models to detect and locate fine-grained details within images.

Click to read more →
ACL 2025

Hybrid Graphs for Table-and-Text based Question Answering using LLMs

A comprehensive framework combining tabular and textual information through hybrid graph structures for improved QA performance.

Click to read more →
NAACL 2025

HOLMES: Hyper-Relational Knowledge Graphs for Multi-hop Question Answering using LLMs

A novel approach using hyper-relational knowledge graphs and query-aligned schemas for accurate multi-hop question answering with LLMs.

Click to read more →
ACL 2024