Pon Rahul Murugaraj

I’m a PhD candidate at the Indian Institute of Management Bangalore, specializing in Information Systems.
My work spans efficient machine learning, edge AI, and embedded systems alongside the evaluation of large language models.

Let’s connect!

What I work on

Systems & Runtimes

Edge AI & Systems

Linux kernel bring-up, neural runtimes (OpenVINO, Vulkan, ROCm, Hexagon), and model compression for real-time edge devices.

Linux Kernel OpenVINO Hexagon

Reliability & Metrics

LLM Evaluation

Measuring factuality, semantic uncertainty, drift, and reliability in large language models to avoid costly human oversight.

Factuality Stochasticity Entropy

Information Retrieval

Search & Retrieval

Hybrid lexical and dense vector search, query understanding, expansion, and cross-encoder re-ranking across enterprise corpora.

BM25 + HNSW RAG Lucene

Streaming & Pipelines

Industrial AI

Compact decision models on resource-constrained hardware, high-throughput streaming pipelines, and drift detection.

Apache Kafka Spark Edge ML

Featured Projects

A selection of my flagship open-source work across embedded Linux, AI behavior evaluation, and hardware-accelerated information retrieval.

Flagship Systems

HP OmniBook 5 Linux Bring-Up

Snapdragon X2 Glymur · qcom-next

Device trees and layered kernel patches on Qualcomm’s upstream tree, enabling working display, GPU, Wi-Fi, Bluetooth, and thermals on real hardware.

GitHub → Details →

AI Evaluation

dow — Drift Observation Workbench

“Git for AI Behavior” · Attribution

Version your complete inference spec, then measure semantic drift, stability, and regressions with causal attribution down to config parameters.

GitHub → Details →

Accelerated IR

LISA — Literature Intelligence

3.3× / 11× on Intel OpenVINO

Hybrid search engine fusing lexical (BM25), dense, and graph retrieval with cross-encoder re-ranking for sentence-level evidence verification.

GitHub → Details →

Explore all projects