Aug 2023 – Apr 2024 · 8 mos · Bengaluru, Karnataka, India · On-site

• Left Google Research (Now called GoogleMind) to start Ananas Labs - an AI R&D startup with the ambition of building something akin to OpenAI or DeepMind, but from India. Led a small but passionate team of interns and moonlighters to develop a novel Indic language architecture, achieving promising results. Unfortunately, limited funding led us to pause operations. This chapter isn’t over—I will be back, stronger.

Nov 2022 – Present · 3 yrs 4 mos

Sep 2021 – Present · 4 yrs 6 mos · Hyderabad, Telangana, India

Nov 2020 – Present · 5 yrs 4 mos

Oct 2019 – Aug 2023 · 3 yrs 10 mos · Bangalore

Jun 2018 – Aug 2019 · 1 yr 2 mos

Apr 2017 – Jun 2018 · 1 yr 2 mos

Mar 2015 – Apr 2017 · 2 yrs 1 mo · Bengaluru Area, India

• I lead Multimedia Research Group at Snapdeal that is a group of immensely talented and hardworking researchers and engineers. The primary mandate of the team is to find Machine Learning solutions to seemingly hard challenges that Snapdeal faces as part of its business and operations.
• 1) New listings validation and standardisation: Amongst its many wins, the biggest one for MRG probably is the complete automation of new listings process. Validating and standardising new listings has been a pain point for e-commerce companies, particularly the ones that function as a marketplace. Most companies have to rely on human labor for this task that is not only costly but also time consuming and difficult to scale. Given its prowess in Machine Learning and Deep Learning (to be more specific), MRG has developed a state-of-the-art system that does completely automatic validation and standardisation of new listings (both text and image). The system is not only cutting edge in terms of ML algorithms and DL networks it uses for the task but also an engineering marvel for being able to achieve massive throughput without using too much hardware.
• 2) Image-based product search: Or more popularly known as Snap-n-Search is a new and intuitive experience for fashion shopping. As part of this project, we advanced the science behind the algorithms that powered Fashiate, in addition to making them work at scale. Unlike other companies, entire research and developement was done in-house by MRG.

Dec 2014 – Mar 2015 · 3 mos · Bengaluru Area, India

• Fashiate is a new and intuitive experience for fashion shopping. Fashiate is for those times when you exactly know what you want, for those times when you have no idea what you want and for all times in between! You can be rest assured your fashiat-ing experience will be enjoyable and leave you uber-satisfied.
• When you are out and about and spot something you like, there is no way to find things that look ‘like that’. Well, not anymore! You can just take a picture and Fashiate will help you find things like that. Naturally, the picture of your favorite product can come from anywhere - movie posters, a magazine ad, a product catalog or anywhere else from the wide web!
• Want to find more pocket-friendly variants of out-of-budget products? Fashiate can find that too in no time. Not only that, you can refine your search based on patterns, colors and styles.

Mar 2014 – Dec 2014 · 9 mos · Bengaluru Area, India

• Worked and successfully led teams for various Computer Vision/Machine Learning projects including
• 1) Best-in-class adult content classification in images and videos
• 2) Very large scale face recognition system capable of accurately recognizing across tens of thousands of celebrities
• 3) Video metadata enrichment by processing pixel data to identify visual concepts and celebrities
• 4) Computational image aesthetics for face portraits and general images
• 5) Article-to-video recommendations

Mar 2012 – Mar 2014 · 2 yrs · Bengaluru Area, India

• Worked and/or led teams for various Computer Vision/Machine Learning projects including
• 1) Best-in-class highly efficient and scalable face identification system
• 2) Robust and efficient face detection system
• 3) Blur-aware photography for mobile phones
• 4) Novel visually-driven CAPTCHAs
• 5) Automatics tagging of images for visual concepts
• 6) Identifying relevant business images for locals
• 7) Automatic detection and identification of text in images

Jun 2010 – Mar 2012 · 1 yr 9 mos · Notre Dame, Indiana

• As a Research Assistant Professor, I taught a graduate-level course, co-advised two Computer Science PhD students, served on M.S. and PhD committees of several students, in addition to making significant contributions to the research projects that dealt with exploring new avenues of biometrics research. In particular, we proposed a novel algorithm for matching faces across real plastic surgeries, a facial mark based approach to distinguish between identical twins and a machine learning approach to match faces across variations in pose, illumination and resolution. The plastic surgery work received extensive media coverage with articles in Wall Street Journal, New Scientist, Indian Express and Zee News.

Jan 2008 – Jun 2010 · 2 yrs 5 mos · Reston, Virginia

• I worked as a Research Scientist with the Center for Video Understanding Excellence (CVUE) group at Object Video. During my stay at Object Video, I made significant contributions in the fields of object recognition, illumination modeling, geo-registration, object detection and tracking, motion estimation, camera calibration and several other challenging computer vision problems and machine learning problems. I procured funding/managed the following projects as principal investigator:
• 1) A Unified Framework for False Alarm Reduction using Scene Context from Airborne Sensors, a DARPA funded project ($1.25M) that provides detection, tracking and classification capabilities for ground station processing of ARGUS sensor data.
• 2) Dismount Tracking, a DARPA funded project ($0.75M) that provides dismount tracking capabilities and 3D structure inference from airborne sensors.
• 3) Multi-Modal Knowledge Acquisition from Documents , an ONR funded STTR project to develop a novel theoretical framework to align text and images.
• I was also a key contributor in the following projects
• 1) Future Naval Capabilities, an ONR funded project ($3.9M) that developed Maritime video analysis for ports.
• 2) Wide Area Motion Deblurring, an STTR Phase II funded by DARPA ($0.75M).

Jan 2006 – Jan 2007 · 1 yr · Hawthorne, New York

• I worked on the following projects in the Exploratory Computer Vision Group (ECVG) with Dr. Nalini Ratha and Dr. Ruud Bolle as my mentors.
• 1) Backward-compatible revocable representations for face: I propose algorithms for user-
• specific photo-metric and structural distortions to obtain secure backward-compatible face representations. The backward-compatibility ensures that any commonly available face recognition
• system can be used to match transformed secure face.
• 2) Gradient-based local texture characterization of fingerprints: I develop a gradient histogram based approach to characterize local fingerprint texture. Such a characterization is useful to improve the performance of purely minutiae-based matching approaches. The characterization can also be used as an independent feature to index fingerprints on a large scale.
• 3) Cohort analysis for biometric matching: I propose algorithms that make use of the often ignored non-matching templates to improve the matching performance. For each enrolled subject, its cohort (similar identities) is identified based on a simple score based selection criterion. The similarity score of a query template is estimated using its similarity not only with the claimed identity but also with its cohort.
• 4) Large scale indexing and identification of fingerprints: We build fingerprint identification system that can match a query against a database of millions of fingerprints in real time. Indexing fingerprints using suitable features helps in performing such a task without explicitly comparing the query with each enrolled print.