Peer-reviewed research at the intersection of knowledge graphs, retrieval augmentation, and large language models.
ICOSST_2024
Published Paper
IEEE · ICOSST 2024PEER-REVIEWEDDECEMBER 2024
To Enhance Graph-Based Retrieval-Augmented Generation (RAG) with Robust Retrieval Techniques
M. Rani, B. K. Mishra, D. Thakker, and M. N. Khan —
18th International Conference on Open-Source Systems and Technologies (ICOSST), pp. 1–6, IEEE, Dec. 2024.
In collaboration with the University of Hull, UK.
This paper proposes enhancements to graph-based RAG pipelines by introducing robust retrieval strategies that improve
the relevance and accuracy of retrieved context in large language model applications. The work addresses key limitations
of vanilla vector-similarity retrieval by leveraging graph traversal and hybrid indexing techniques to improve
answer grounding in domain-specific knowledge bases.
Venue
ICOSST 2024
Publisher
IEEE
Date
December 2024
Pages
1–6
Collaboration
University of Hull, UK
Topic
Graph-based RAG
RESEARCH_INTERESTS
Areas of Focus
🔗
Graph-based RAG
Knowledge graph construction and traversal for improved context retrieval in LLM pipelines. Hybrid graph + vector indexing strategies.
⚡
LLM Inference Optimization
Quantization, speculative decoding, tensor parallelism and kernel fusion for low-latency LLM serving at scale on GPU hardware.
🤖
Multi-Agent AI Systems
Designing orchestration frameworks for collaborative AI agents in complex task environments. Tool use, planning, and memory architectures.
👁
Computer Vision & Medical AI
Attention-based architectures for medical imaging, edge-deployed vision models, and real-time object detection in resource-constrained environments.
🏗
MLOps at Scale
Reproducible, observable, and reliable ML systems in production — spanning GPU cluster management, CI/CD for ML, and model lifecycle governance.
🌱
Generative Models
Fine-tuning and evaluation of diffusion models and large language models for domain-specific generation tasks including code, speech, and images.