cv | Ajay Narayanan Sridhar

Basics

Name	Ajay Narayanan Sridhar
Label	Ph.D. student in Electrical and Computer Engineering at Georgia Tech
Email	narayananajay99@gmail.com
Url	https://ajaynarayanan.github.io
Summary	Research in multimodal AI, retrieval-augmented systems, efficient on-device inference, and contextual bandits for LLM routing. Advised by Prof. Vijaykrishnan Narayanan; expected graduation in May 2027.

Education

2027.05 (Expected)

Atlanta, GA
Ph.D.

Georgia Institute of Technology

Electrical and Computer Engineering
- Advisor: Prof. Vijaykrishnan Narayanan
2024.08

University Park, PA
M.S.

The Pennsylvania State University

Computer Science and Engineering
- Advisor: Prof. Vijaykrishnan Narayanan
2020.05

Chennai, India
Bachelor of Technology

IIITDM Kancheepuram

Computer Engineering (Honours), specialization in AI and ML

Work

2026.05 - 2026.08

San Francisco, CA
PhD Machine Learning Engineer Intern

Dropbox
- Developed a dense multimodal video retrieval system using CLIP frame embeddings for image- and text-based search at precise timestamps, improving NDCG@10 by 30% over production episode-level retrieval.
- Won Best Technical Execution and Most Likely to Ship at the Dropbox Emerging Talent Hackathon 2026 for an end-to-end agentic AI system that triaged customer issues, enriched diagnostics, routed bugs to code owners, and generated JIRA tickets and AI-generated PRs for human review.
2024.08 - Present

Atlanta, GA
Graduate Research Assistant

Microsystems Design Lab
- Investigating multimodal retrieval-augmented systems that combine learned embeddings, compact memory representations, and LLM reasoning for question answering and contextual recall over long-horizon user data.
- Developing EdgeVQA with QLoRA fine-tuning of Qwen2.5-VL-3B on VizWiz-VQA and INT8/INT4 (AWQ) quantization for efficient on-device visual question answering, extending NaviSense toward real-time, privacy-preserving mobile deployment.
2022.04 - 2022.07

Chennai, India
Software Engineer

Navis LLC
- Designed and developed an Apache Kafka-based, scalable, configuration-driven RESTful machine learning platform for predictive and prescriptive use cases in shipping terminal operations.
- Implemented missing-data handling and validation in analytics workflows, delivering 47% faster execution and fewer production failures.
2020.08 - 2022.03

Chennai, India
Associate Software Engineer

Navis LLC
- Optimized batch analytics through algorithmic refactoring and pipeline redesign, reducing runtime by 31% for customer-facing reporting workflows.
- Implemented ETL failure detection and diagnostics, reducing root-cause analysis time by 40%.

Projects

Correlation-Aware Contextual Bandits

Applications in LLM routing.
- Formulated LLM routing as a contextual bandit with correlated arms and surrogate rewards; designed correlation-aware algorithms with regret guarantees robust to surrogate errors while improving regret when surrogates are informative.
- Demonstrated up to approximately 15% higher utility across multiple LLM routing benchmarks versus strong online learning baselines and competitive state-of-the-art routing methods.
NaviSense

Multimodal AI workflow for real-world task guidance.
- Engineered a real-time multimodal pipeline combining LiDAR depth, visual inputs, and LLM reasoning for grounded, step-by-step task guidance in open-world environments.
- Deployed LLaMA 3.2 1B Instruct on-device via ExecuTorch with 4-bit quantization, balancing latency, memory, and response quality for mobile inference.

Skills

	LLMs and Generative AI
	Transformers
	RAG
	Question answering
	Model routing and selection
	Prompt design
	Grounding and verification

	Machine Learning
	PyTorch
	TensorFlow
	Fine-tuning
	Training and evaluation
	Benchmarking

	Programming
	Python
	Java
	C/C++
	Swift

	Data and Systems
	OpenSearch
	MySQL
	PostgreSQL
	Redis
	REST APIs
	Apache Kafka
	Spring Boot

	Cloud and DevOps
	Docker
	Kubernetes
	AWS
	Azure
	GCP
	Git

Awards

2026

Best Technical Execution and Most Likely to Ship

Dropbox

Dropbox Emerging Talent Hackathon 2026.
AI Societal Impact Award

Penn State

For contributions to research for people with visual impairments.
c0mpiled Hackathon Winner

c0mpiled

For DoomScholar, an anti-doomscrolling app.
2025

Best Audience Choice Poster Award

ACM SIGACCESS

For NaviSense at ASSETS '25.
Dr. Tse-Yun Feng Graduate Student Award and Scholarship

Penn State

For academic and research excellence.
Undergraduate Merit Prize

IIITDM Kancheepuram

Honours student.

Interests

	Research
	Multimodal retrieval and memory
	Retrieval-augmented generation
	Visual accessibility
	On-device AI
	Contextual bandits and LLM routing
	Model optimization

	Outside research
	Cycling
	Working out
	Creative accessibility technologies

2026

NeurIPS

Correlation-Aware Contextual Bandits with Surrogate Rewards for LLM Routing

Ajay Narayanan Sridhar, Ronak Singh, Mehrdad Mahdavi, and Vijaykrishnan Narayanan

arXiv preprint arXiv:2607.09015, 2026

Accepted to NeurIPS 2026 (main conference)

Bib HTML

@article{sridhar2026correlation,
  title = {Correlation-Aware Contextual Bandits with Surrogate Rewards for LLM Routing},
  author = {Narayanan Sridhar, Ajay and Singh, Ronak and Mahdavi, Mehrdad and Narayanan, Vijaykrishnan},
  journal = {arXiv preprint arXiv:2607.09015},
  year = {2026},
  note = {Accepted to NeurIPS 2026 (main conference)},
  eprint = {2607.09015},
  archiveprefix = {arXiv},
  url = {https://arxiv.org/abs/2607.09015},
}

NeurIPS

Conditional Optimal Bridge for Riemannian Activation Steering

Seyed Arshan Dalili, Ajay Narayanan Sridhar, Vijaykrishnan Narayanan, and Mehrdad Mahdavi

arXiv preprint arXiv:2607.10517, 2026

Accepted to NeurIPS 2026 (main conference)

Bib HTML

@article{dalili2026conditional,
  title = {Conditional Optimal Bridge for Riemannian Activation Steering},
  author = {Dalili, Seyed Arshan and Narayanan Sridhar, Ajay and Narayanan, Vijaykrishnan and Mahdavi, Mehrdad},
  journal = {arXiv preprint arXiv:2607.10517},
  year = {2026},
  note = {Accepted to NeurIPS 2026 (main conference)},
  eprint = {2607.10517},
  archiveprefix = {arXiv},
  url = {https://arxiv.org/abs/2607.10517},
}

2025

ASSETS

NaviSense: A Multimodal Assistive Mobile Application for Object Retrieval by Persons with Visual Impairment

Ajay Narayanan Sridhar, Fuli Qiao, Nelson Daniel Troncoso Aldas, Yanpei Shi, Mehrdad Mahdavi, Laurent Itti, and Vijaykrishnan Narayanan

In Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS ’25), Oct 2025

Poster Award Bib Video

Best Audience Choice Poster Award at ASSETS ’25.

@inproceedings{sridhar2025navisense,
  title = {NaviSense: A Multimodal Assistive Mobile Application for Object Retrieval by Persons with Visual Impairment},
  author = {Narayanan Sridhar, Ajay and Qiao, Fuli and Troncoso Aldas, Nelson Daniel and Shi, Yanpei and Mahdavi, Mehrdad and Itti, Laurent and Narayanan, Vijaykrishnan},
  booktitle = {Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '25)},
  year = {2025},
  month = oct,
  address = {Denver, CO, USA},
  publisher = {ACM},
}

RSD ’26

Evaluating Large Language Models on Rare Disease Diagnosis: A Case Study using House M.D.

Arsh Gupta, Ajay Narayanan Sridhar, Bonam Mingole, and Amulya Yadav

2025

Accepted to the AAAI 2026 Workshop on Responsible Synthetic Data (RSD’26)

Bib HTML

@misc{gupta2025evaluating,
  title = {Evaluating Large Language Models on Rare Disease Diagnosis: A Case Study using House M.D.},
  author = {Gupta, Arsh and Narayanan Sridhar, Ajay and Mingole, Bonam and Yadav, Amulya},
  year = {2025},
  eprint = {2511.10912},
  archiveprefix = {arXiv},
  url = {https://arxiv.org/abs/2511.10912},
  note = {Accepted to the AAAI 2026 Workshop on Responsible Synthetic Data (RSD'26)},
}

ISVLSI

Insect Agent: Improving Insect Recognition via Dynamic Knowledge Augmentation Using Multimodal Large Language Models

Shu Zhao^*, Ajay Narayanan Sridhar^*, Harland Patch, and Vijaykrishnan Narayanan

In 2025 IEEE Computer Society Annual Symposium on VLSI (ISVLSI), 2025

* indicates equal contribution.

Bib Blog Code

@inproceedings{insectAgent,
  title = {Insect Agent: Improving Insect Recognition via Dynamic Knowledge Augmentation Using Multimodal Large Language Models},
  author = {Zhao, Shu and Narayanan Sridhar, Ajay and Patch, Harland and Narayanan, Vijaykrishnan},
  booktitle = {2025 IEEE Computer Society Annual Symposium on VLSI (ISVLSI)},
  year = {2025},
  organization = {IEEE},
  note = {* indicates equal contribution.},
}

2024

CVPR W

A Generative Exploration of Cuisine Transfer

Philip Wootaek Shin^*, Ajay Narayanan Sridhar^*, Jack Sampson, and Vijaykrishnan Narayanan

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024

* indicates equal contribution.

Bib

@inproceedings{shin2024generative,
  title = {A Generative Exploration of Cuisine Transfer},
  author = {Shin, Philip Wootaek and Sridhar, Ajay Narayanan and Sampson, Jack and Narayanan, Vijaykrishnan},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
  pages = {3732--3740},
  year = {2024},
  note = {* indicates equal contribution.},
}

MM Asia

QoS-Diff: Adaptive Auto-tuning framework for Low-latency Diffusion Model Inference

Pingyi Huo^*, Ajay Narayanan Sridhar^*, Md Fahim Faysal Khan, Kiwan Maeng, and Vijaykrishnan Narayanan

In Proceedings of the 6th ACM International Conference on Multimedia in Asia, 2024

* indicates equal contribution.

Bib

@inproceedings{lu2023rl,
  title = {QoS-Diff: Adaptive Auto-tuning framework for Low-latency Diffusion Model Inference},
  author = {Huo, Pingyi and Narayanan Sridhar, Ajay and Khan, Md Fahim Faysal and Maeng, Kiwan and Narayanan, Vijaykrishnan},
  booktitle = {Proceedings of the 6th ACM International Conference on Multimedia in Asia},
  year = {2024},
  note = {* indicates equal contribution.},
}

2019

ICCV W

Feature Ensemble Networks with Re-Ranking for Recognizing Disguised Faces in the Wild

Arulkumar Subramaniam^*, Ajay Narayanan Sridhar^*, and Anurag Mittal

In The IEEE International Conference on Computer Vision (ICCV) Workshops, Oct 2019

Oral presentation; * indicates equal contribution.

Bib HTML PDF Slides

@inproceedings{Subramaniam_2019_ICCV_Workshops,
  title = {Feature Ensemble Networks with Re-Ranking for Recognizing Disguised Faces in the Wild},
  author = {Subramaniam, Arulkumar and Narayanan Sridhar, Ajay and Mittal, Anurag},
  booktitle = {The IEEE International Conference on Computer Vision (ICCV) Workshops},
  month = oct,
  year = {2019},
  note = {Oral presentation; * indicates equal contribution.},
}

Basics

Education

Ph.D.

Georgia Institute of Technology

Electrical and Computer Engineering

M.S.

The Pennsylvania State University

Computer Science and Engineering

Bachelor of Technology

IIITDM Kancheepuram

Computer Engineering (Honours), specialization in AI and ML

Work

PhD Machine Learning Engineer Intern

Dropbox

Graduate Research Assistant

Microsystems Design Lab

Software Engineer

Navis LLC

Associate Software Engineer

Navis LLC

Projects

Correlation-Aware Contextual Bandits

Applications in LLM routing.

NaviSense

Multimodal AI workflow for real-world task guidance.

Skills

Awards

Best Technical Execution and Most Likely to Ship

Dropbox

Dropbox Emerging Talent Hackathon 2026.

AI Societal Impact Award

Penn State

For contributions to research for people with visual impairments.

c0mpiled Hackathon Winner

c0mpiled

For DoomScholar, an anti-doomscrolling app.

Best Audience Choice Poster Award

ACM SIGACCESS

For NaviSense at ASSETS '25.

Dr. Tse-Yun Feng Graduate Student Award and Scholarship

Penn State

For academic and research excellence.

Undergraduate Merit Prize

IIITDM Kancheepuram

Honours student.

Interests

Publications

2026

2025

2024

2019