Mahesh Sathiamoorthy

Mahesh Sathiamoorthy

About

Detail

Co-Founder
Mountain View, California, United States

Timeline


work
Job
school
Education
auto_stories
Publication

Résumé


Jobs verified_user 0% verified
  • Bespoke Labs
    Co-Founder
    Bespoke Labs
    Apr 2024 - Current (2 years 3 months)
    We work on post-training. Current focus is on using RL for post-training agents. Get out of the prompt hell! We created open-thoughts.ai along with members in open-source community: we have open-sourced some of the best reasoning datasets (cited by Nvidia, Microsoft) and trained OpenThinker2-32B which beats DeepSeek's 32B distilled model. For data curation, we have https://github.com/bespokelabsai/curator.
  • Google Deepmind
    Staff Software Engineer
    Google Deepmind
    Oct 2020 - Dec 2023 (3 years 3 months)
    LLMs ↔ Large Recommender Models.
  • Google
    Senior Software Engineer
    Google
    Aug 2017 - Oct 2020 (3 years 3 months)
    TPUs + Neural recommenders @ Google Brain.
  • Google
    Software Engineer
    Google
    Jan 2014 - May 2016 (2 years 5 months)
    I increased the utilization of flash at Google by *several orders of magnitude*. This required building pipelines to characterize workloads, implement optimization algorithms that decide which workloads to put on flash (and for how long), and building dashboards to help internal customers understand their workloads.
  • TURN
    Big Data Intern
    TURN
    Sep 2013 - Nov 2013 (3 months)
  • Symantec
    Intern (Symantec Research Labs)
    Symantec
    May 2013 - Aug 2013 (4 months)
  • General Motors
    Visiting Scholar
    General Motors
    May 2011 - Aug 2011 (4 months)
    Visiting Scholar at General Motors R&D working on Vehicular Networks
  • University of Southern California
    PhD Student
    University of Southern California
    Aug 2008 - Dec 2013 (5 years 5 months)
  • University of Southern California
    Intern
    University of Southern California
    May 2007 - Jul 2007 (3 months)
    Worked on a research project titled "Compressed Wireless Sensing"
  • Nanyang Technological University Singapore
    Intern
    Nanyang Technological University Singapore
    May 2006 - Jul 2006 (3 months)
    Worked on various topics related to Speech Recognition and built a Speech Recognizer
Education verified_user 0% verified
  • University of Southern California
    M.S & PhD, Computer and Wireless Networking
    University of Southern California
    Jan 2008 - Dec 2013 (6 years)
  • Indian Institute of Technology Kharagpur
    B.Tech(H, Electronics and Electrical Communication Engg
    Indian Institute of Technology Kharagpur
    Jan 2004 - Dec 2008 (5 years)
Projects (professional or personal) verified_user 0% verified
  • X
    XORing Elephants: Novel Erasure Codes for Big Data
    Xorbas is modification of Hadoop's RAID module that incorporates a new set of regenerating erasure codes called Locally Repairable Codes (LRC).
Publications verified_user 0% verified
  • I
    Helper Node Allocation Strategies for Content Dissemination in Intermittently Connected Mobile Networks
    IEEE International Conference on Sensing Communication and Networking SECON
    Jan 2014
    We formulate and address the fundamental problem of allocating helper nodes in disseminating multiple content in an intermittently connected mobile network. We consider and solve two variations of the problem - one in which the goal is to maximize the expected demands satisfied and another in which the goal is to minimize the time taken to disseminate the contents. Further, we study this problem in a game theoretic perspective and show the effect of selfishness on the utility of the system.
  • V
    Dynamic Online Storage Allocation for Multi-Content Dissemination in Two-Tier Hybrid Mobile Vehicular Networks
    Vehicular Network ConferenceIEEE
    Nov 2013
    The fleet of smart mobile devices in every aspect of daily life, has been raised serious concerns about the bandwidth availability within the current cellular structure. To address such concerns in urban areas, we present a two-tier hybrid mobile net- work. The two tiers are data plane consists of store and forward routing through an intermittently connected mobile network, and the control plane consists of an always-on infrastructure- based wireless network. With the use of such structure we can avoid bandwidth limitation while still maintaining the central- ized controlling aspect in data delivery. For such a structure, we formulate and address,from a theoretical perspective, the fundamental problem of how to dynamically allocate storage i
  • t
    XORing Elephants: Novel Erasure Codes for Big Data
    th International Conference on Very Large Data Bases VLDB
    Jan 2013
    Xorbas is our Hadoop version that implements a new set of regenerating codes called Locally Repairable Codes (LRC). It is built on top of Facebook's Hadoop system running HDFS-RAID, and thus can support Reed Solomon and XOR codes in addition to LRCs. Abstract: Distributed storage systems for large clusters typically use replication to provide reliability. Recently, erasure codes have been used to reduce the large storage overhead of three-replicated systems. Reed-Solomon codes are the standard design choice and their high repair cost is often considered an unavoidable price to pay for high storage efficiency and high reliability. This paper shows how to overcome this limitation. We present a novel family of erasure codes that are efficientl
  • V
    Minimum Latency Data Diffusion in Intermittently Connected Mobile Networks
    Vehicular Technology Conference VTC Spring IEEE th
    May 2012
    We consider the problem of diffusing cached content in an intermittently connected mobile network, starting from a given initial configuration to a desirable goal state where all nodes interested in particular contents have a copy of their desired contents. The goal is to minimize the time taken for the diffusion process to terminate at a goal state. Due to bandwidth and storage constraints, whenever two nodes encounter each other, they must decide which content if any to transfer to each other. While most prior work on this topic has focused on practically realizable heuristics for this problem, we take a more formal approach. Our main contribution is to show that, assuming global state information is available, this problem can be formula
  • I
    Distributed Storage Codes Reduce Latency in Vehicular Networks
    IEEE InfocomMini
    Jan 2012
    We investigate the benefits of distributed storage using erasure codes for file sharing in vehicular networks through realistic trace-based simulations. We find that coding offers substantial benefits over simple replication when the file sizes are large compared to the average download bandwidth available per encounter. Our simulations, based on a large real vehicle trace from Beijing combined with a realistic radio link quality model for a IEEE 802.11p dedicated short range communication (DSRC) radio, demonstrate that coding provides significant cost reduction in vehicular networks.
This is a community-created genome.