cv
Downloadable CV. Replace the PDF with your latest version when ready.
General Information
| Full Name | Sai Narayan Sundaresan |
| saisundaresan01@gmail.com | |
| Website | https://saisundaresan01.github.io |
| Google Scholar | https://scholar.google.com/citations?user=icwyFlQAAAAJ&hl=en |
| https://www.linkedin.com/in/sai-narayan-sundaresan/ |
Education
-
2019 -- 2024 Dual Degree (B.Tech, M.Tech), Industrial and Systems Engineering
Indian Institute of Technology Kharagpur, India - {"GPA"=>"9.15/10.00"}
- {"Micro-specialization"=>"Artificial Intelligence and Applications"}
Research Experience
-
Jun 2024 -- Present Research Associate – Adobe Inc. (Systems and Insights Group)
Adobe Inc. - {"Mentors"=>"Dr. Subrata Mitra, Dr. Atanu Sinha, Dr. Shiv Saini"}
- Worked on efficient systems for LLM serving and video generation via caching techniques
- Published 2 papers and filed 3 patents; integrated innovations in 2 products
- Selected for a session on world models at Adobe Tech Summit 2025
-
Jan 2024 -- May 2024 Research Intern – Sarvam.AI
Sarvam.AI - {"Mentors"=>"Prof. Pratyush Kumar, Dr. Vivek Raghavan"}
- Built end-to-end tool creating guided audio lessons using custom speech pipelines
- Deployed efficient ASR service using NVIDIA RIVA for low-resource languages
- Trained expressive TTS models based on StyleTTS2
-
May 2023 -- Aug 2023 Research Intern – Adobe Inc. (BigData Intelligence Lab)
Adobe Inc. - {"Project"=>"Brand-guided Poster Generation"}
- Built design creation tool using Custom Diffusion with product and brand references
- Devised optimal multi-concept latent interpolation algorithm
-
Jan 2023 -- Apr 2023 Data Engineering Intern – AI4Bharat, IIT Madras
IIT Madras (AI4Bharat) - {"Mentor"=>"Prof. Mitesh M. Khapra"}
- Collated ASR benchmarks across 12 Indian languages; created 10,736 hours dataset
- Performed alignment with Needleman–Wunsch; finetuned Whisper achieving 4.1 WER reduction
-
May 2022 -- Jul 2022 Machine Learning Intern – Honeywell Technology Solutions Lab
Honeywell Technology Solutions Lab - {"Mentor"=>"Mr. Srikanth Viswaraju"}
- Designed contextual search engine for flight manuals using BERT; 79.38% text / 54.69% tabular
Selected Projects
-
ongoing Efficient Video Generation through Patch Level Caching
- Reuse intermediate states across similar patches via rectified flow interpolation.
-
2025 Efficient LLM Serving for RAG (Cache-Craft)
- Chunk-cache decomposition and reuse; ~2.7× prefill latency reduction.
-
2025 Heterogeneity-Aware User Behaviour Prediction (HeTLM)
- Clusterwise training showing small LMs better model subjective browsing behaviors.
Activities and Achievements
-
2025 - Mentored 6 undergraduate interns and collaborated with 1 PhD intern at Adobe
- Best paper and merit awards at ORSI '24 and ICLST '24
- Branch Rank 1 in Quality Engineering Design and Manufacturing (highest GPA)
- Certificate of Merit for top rank in Computer Science, AISSCE 2019
Research Interests
-
ML + Systems for generative models
- Efficient LLM serving and caching for RAG
- Efficient video generation via computation reuse
Skills
- {"Programming"=>"Python, C++, C, Triton"}
- {"Frameworks"=>"vLLM, Transformers, Diffusers, PyTorch, Git"}