Lavanya Sita Tekumalla

    Industry: Over 7 years - Amazon, InMobi, Myntra  

    Masters: Computer Graphics   ( 2002 - 2005 )
    School of Computing, University of Utah  

    Ph.D: Machine Learning   ( 2011 - 2016 )
    Dept. of Computer Science and Automation(CSA)
    Indian Institute of Science(IISc), Bangalore


About Me || Research || Publications || Industry || Education

About Me

    I am currently an independant AI/ML consultant. Until recently, I have been a Machine Learning Scientist at Amazon for the Core Machine Learning Team that solves a range of problems partnering with various Amazon teams to drive customer Impact. I had particularly worked on problems from the Amazon Fashion Vertical.

    I had completed a PhD from IISc, where I performed most of my doctoral research @ the Machine Learning Lab, CSA, IISc. I was advised by Dr. Chiranjib Bhattacharya. Prior to that, I was advised by Dr. Indrajit Bhattacharya. My research focus during PhD has been on Bayesian Non-parametric modeling, Probabilistic Graphical models and their applications. I am also interested in approaches for scaling inference algorithms and recently deep learning based techniques. So far, I have worked on probabilistic models for the e-commerce domain, systems domain and Natural Language Processing.

    Prior to joining PhD, I had worked in the industry in a variety of roles at, and InMobi(Bangalore). During my PhD, I have worked closely with NetApp(Bangalore) on designing machine learning models for mining block I/O traces and have also interned at Xerox Research Center(Bangalore), where I worked with modeling with Copulas.

    I had also worked with Prof. Elaine Cohen (Geometric Design and Computation Group) in the Area of Geometric Modeling with B-Splines, during my Masters at University of Utah. My work involved finding novel algorithms for fitting B-Spline surfaces from noisy and incomplete point cloud data.

    Note: Recently, sevaral students have reached out to me for mentoring and interview preparation tips. I am curating some interview questions on Feel free to check these out.


  • Product Size Recommendation for E-commerce (@Amazon) :

    -- Bayesian Models For Product Size Recommendation, WWW 2018, Vivek Sembium, Rajeev Rastogi, Lavanya Tekumalla, Atul Saroop.

  • Copula Based Modeling for Mixed Discrete Continuous Data (In collaboration with Xerox Research Center India-XRCI) :

    -- "Vine copulas for mixed data : multi-view clustering for mixed data beyond meta-Gaussian dependencies", Lavanya Sita Tekumalla, Vaibhav Rajan, Chiranjib Bhattacharyya, Machine Learning Journal, 2017, 10.1007/s10994-016-5624-2. (pdf)

    -- Patent Filed: Methods and Systems For Predicting the Health Condition Of a Human Subject, Inventors: Lavanya Tekumalla, Vaibhav Rajan, 20140865US01.

  • Bayesian Non-parametric modeling for systems data (In collaboration with Netapp, Bangalore) :
    Existing techniques for cache Preloading are based on analyzing short range block access level correlations. We address the novel problem of bulk preloading of Gigabytes of data at a time, minutes or hours in advance by reducing an I/O trace to a a sequence of count vectors and proposing novel BNP techniques for multivariate count data analysis (a previously unexplored topic). This is ongoing interdisciplinary work.

    -- Copula-HDP-HMM: Non-parametric models with Copulas for I/O Efficient Bulk Cache Preloading, Lavanya Sita Tekumalla, Chiranjib Bhattacharyya, Accepted at SIAM Data Mining conference, 2016.

    -- Mining Block I/O Traces for Cache Preloading with Sparse Temporal Non-parametric Mixture of Multivariate Poisson, Lavanya Sita Tekumalla, Chiranjib Bhattacharyya, SIAM Data Mining conference, Vancouver, 2015 (pdf)

    -- Honourable mention @ Xerox Research Center India Colloquium 2016.

    -- Invited Talk May 2015 @ Facebook AI Research, Palo Alto, USA: 'Statistical Models for Bulk Cache Preloading', Lavanya Sita Tekumalla

    -- Invited Talk April 2015 @ NetApp University Day, Bangalore, India: 'Mining Block I/O traces for Bulk Cache Preloading', Lavanya Sita Tekumalla

  • Multi-level Non-parametric Admixture Models - nHDP (With collaborators from IRL, Bangalore) :

    -- "Multilevel Nonparametric Admixture Modeling with Nested Hierarchical Dirichlet Process", Lavanya Tekumalla*, Priyanka Agarwal*, and Indrajit Bhattacharya. (* equal contribution authors), arXiv:1508.06446, 2015 (pdf)

    -- "Nested Hierarchical Dirichlet Process for Nonparametric Entity-Topic Analysis", Priyanka Agrawal*, Lavanya Tekumalla*, and Indrajit Bhattacharya. European Conference of Machine Learning 2013, Prague. (* equal contribution authors) (pdf)

    -- Invited Talk 2014 @ Xerox Research Labs India, Bangalore, India: "Nested Hierarchical Dirichlet Process for Nonparametric Entity-Topic Analysis", Lavanya Tekumalla

  • Masters Thesis: Reverse Engineering Point Clouds to Obtain Tensor Product B-Spline Surfaces:

    --"Smoothing Space Curves with the MLS Projection", Lavanya Sita Tekumalla, Elaine Cohen, Proceedings of Symposium on Geometric Modeling, Vizualization and Graphics, Salt Lake City, July 2005. (pdf).

    --"Hole-Filling Algorithm for Triangular Meshes"(41 citations), Lavanya Sita Tekumalla, Elaine Cohen, Technical Report, School of Computing, University of Utah, Dec 2004, UUCS-04-019. (pdf)

    --"Reverse Engineering Point Clouds to Fit Tensor Product B-Spline Surfaces by Blending Local Fits, Tech Report", May 2005 (pdf)

    -- Talk : "Reverse Engineering Point Clouds to Trimmed NURBS" , presented as a talk at the conference 'Mathematical Methods for Curves and Surfaces', Oslo, July 2004.


  • PhD in CS : GPA - 7.0/8 : Indian Institute of Science (IISC), Bangalore.

  • (Google Anita Borg Scholar APAC - 2014)

  • MS in CS with Computer Graphics Emphasis: GPA-3.97/4 : School of Computing, University of Utah

  • BE in CS : 89% : Osmania University College of Engineering(Autonomous) - OUCE, Hyderabad

  • (GATE All India Rank 2 , Rank 4 in University)
Other Academic Activities at CSA, IISc
    TA for Algorithms and Programming - ESC 101 (Aug-Dec 2012)
    TA for Linear Algebra Course - E0219 (Aug-Dec 2013)
    TA for Machine Learning - E0270 (Jan-Apr 2014, 2015)
    Reviewer: Robotics and Autonomous Systems (Elsevier Journal), Machine Learning For Big Data (Sadhana Journal)


    I have spent a considerable time in the industry working in a variety of roles that I thoroughly enjoyed.

  • 2017 Jan-: : I have worked for the Core Machine Learning team of Amazon that solves a range of problems partnering with various Amazon teams to drive customer Impact. I have particularly worked on problems from the Amazon Fashion Vertical during this period.

  • 2005 April-2008 Nov: : I have worked for designing and developing software for the Amazon Fulfilment Center. I worked on various algorithms to improve the productivity of the work force though optimal work assignment strategies at the Amazon warehouses.

  • 2009 Jan - 2009 Dec : : is a Bangalore based e-commerce start-up. I worked primarily on software to improve and finetune their order management pipeline for fullfilment.

  • 2010 Jan - 2011 July : InMobi: InMobi is a Bangalore based start-up. InMobi SmartPay is a mobile payment gateway that enables various modes of payments for digital merchandize. I joined InMobi SmartPay at its inception and helped build the codebase from scratch and lead the development effort for expanding this product to sevaral countries.