Shivam Raisharma

Experience

• 

  ESAMYAK SOFTWARE PVT LTD.

  Jun 2018 - Jun 2018

  Summer Intern

  Web development with mssql, Java, html, css

• 

  Thakur College Of Engineering and Technology

  Mar 2019 - Apr 2019

  Student Intern
• 

  NVIDIA

  Dec 2019 - Jan 2020

  Deep Learning Intern

  Team SAE● Primary contributor in enhancing XLNet architecture with data and model parallelism using Megatron-LM. ● Devised 1st model parallel solution for two stream attention module while accounting for recurrence in XLNet, achieving weak scaling efficiency of 76% and beating data parallelism only & inter-layer pipeline parallelism baselines on SQuaD ● Optimized computational bottlenecks, using GPU profilers, and trained LLMs distributed synchronously & asynchronously, across 100s of GPUs with mixed precision data parallelism, docker, slurm & singularity with Nvidia HPC GPU ClusterOther skills pickedup : Nvidia DeepStream, Nvidia TensorRT Show less

• 

  Carnegie Mellon University

  Dec 2020 - Jan 2021

  Research Intern

  - Disentangled spatial representations in Cryo-Electron Tomography (Cryo-ET) data by developing a variational autoencoder (VAE) with a von Mises-Fisher distribution prior, improving the interpretability of latent dimensions.- Demonstrated marginal improvements over standard VAEs with Gaussian priors, highlighting the effectiveness of non-Euclidean priors in capturing structural patterns within Cryo-ET datasets.- Collaborated with a multidisciplinary team to apply advanced machine learning techniques in computational biology, contributing to the lab's ongoing projects in structural pattern mining.- Presented findings in lab meetings, showcasing how disentangled representations can enhance data analysis and structural insights in biological imaging. Show less

• 

  NVIDIA

  Feb 2021 - Aug 2021

  Product Intern Machine Learning

  - Developed a web application for meeting summarization and action item recommendations, powered by a custom-trained BART model on the CNN dataset.- Optimized deployment on NVIDIA Triton, reducing model size by 50% and accelerating inference speed by 1000%, enabling seamless, responsive web interactions.- Integrated NVIDIA Riva’s Automatic Speech Recognition (ASR) for real-time, accurate meeting transcription directly within the app, facilitating text-based summaries.- Enabled Text-to-Speech (TTS) with NVIDIA Riva, allowing users to listen to summarized content and action items, enhancing accessibility and user experience.- Utilized Riva’s Natural Language Processing (NLP) capabilities to automate extraction of key information and action items from transcripts, streamlining meeting documentation.- Added multilingual support via Riva, enabling the web app to process and summarize meetings conducted in various languages, making it adaptable for diverse user groups.- Achieved real-time functionality with Riva's optimization for rapid processing, delivering instant transcription, summarization, and action item recommendations directly in the web interface. Show less

• 

  University of California, Davis

  Apr 2023 - now

  RUBINET Lab under the guidance of Prof Chen-Nee Chuahhttps://www.ece.ucdavis.edu/~chuah/rubinet/people/group.htmlGenAI, MLLMs, Few Shot Learning, Semi Supervised Learning● Alleviating hallucinations in MLLMs like Llava1.5, MiniGPT4, InstructBLIP, GPT4-o, by addressing issues in SFT datasets; developing benchmarks for instruction faithfulness and exposing language bias of MLLMs with Stable Diffusions (ongoing).● Co-leading a $6M cross-campus (UCSF, UCSD, Davis, Emory, Columbia, Davis) RO1 project on WM/GM and neurofibrillary tangle segmentation in Alzheimer's brain tissue [WSIs], using Yolo, SAM, foundation models & LORA. [paper]● (1st Author) Domain Adaptating CLIP using only 10 mins of training & 1% of data; and improving vision-language modality alignment for pathology domain.● Conducted 100 experiments on Few Shot Learning in a data-efficient setting to devise a pipeline achieving over 90% accuracy on images using FSL, foundation models, and pathology-aware post-hoc filtering (Submission to ICCV 2025). ● Developed a deployable Multimodal ML pipeline achieving explainable 90% AUC ROC for early ARDS detection from EHR (tabular) & ventilator waveform data using model calibration for sensitivity and fairness.● Accepted for UCDMC ICU care quality improvement program (1st Author Submission to Nature Scientific Reports).● Accepted poster at Society of Maternal Fetal Medicine and accepted abstract at American Journal of Obstetrics and Gynaecology for readmission risk prediction of preeclampsia and the effects of Social Determinants of Health factors on readmission Show less

  • Graduate Student Researcher

    Dec 2022 - now

  • Graduate Teaching Assistant

    Apr 2023 - Jun 2023
• 

  UC Davis Health

  Jun 2023 - Jun 2024

  Graduate Student Researcher

  ● Shadowed Dr. Jason Adams monthly on both the medical ICU and cardiothoracic ICU services at UC Davis Health, a large, tertiary academic medical center in Sacramento, CA.● Observed bedside procedures in the ICU and open heart surgery in the operating room.● Observed data generation and utilization by multiple clinician types (eg - doctors, nurses, pharmacists, and respiratory therapists) and technology integration in clinician workflows using the Epic electronic health record (EHR) and other non-Epic monitoring and medical device technologies.● Participated in teaching rounds in the ICU and learned complex physiology and pathophysiology principles related to acute respiratory failure and the acute respiratory distress syndrome (ARDS), septic shock including fluid resuscitation and Starling’s law, the effects of vasoactive medications on cardiac and circulatory physiology, the use of mechanical ventilation, and the processof differential diagnosis● Developed a deployable Multimodal ML pipeline achieving explainable 90% AUC ROC for early ARDS detection from EHR (tabular) & ventilator waveform data using model calibration for sensitivity and fairness. Show less