Oct 2022 – Dec 2023 · 1 yr 2 mos · Bengaluru, Karnataka, India · On-site

• 1. Worked on High resolution satellite data(Sentinel, DEM, Internal data sources etc)
• 2. Built capability to understand agricultural landscape in India: https://agri.withgoogle.com/

Aug 2021 – Dec 2021 · 4 mos · Bengaluru, Karnataka, India

• Entrepreneur First (EF) is an international venture capital fund & the world's leading talent investor. EF is backed by Peter Thiel, Reid Hoffman, Greylock Partners, Mosaic Ventures, Lakestar and Founders Fund. With an acceptance rate of 3% each cohort, EF selects 30 to 40 individuals across London, Bangalore, Berlin, Singapore, Paris and Toronto - to help them build and scale deep-tech startups.
• Since 2011, EF cohort members have built companies that have a combined valuation of $4B.

Mar 2020 – Aug 2021 · 1 yr 5 mos · Bengaluru, Karnataka, India

• Implemented text-matching algorithm by giving differential weights to the position of matched characters.
• Implemented vectorized UDFs that reduced the cost and execution time by significant margin.
• Implemented dynamic job orchestration for pipeline driven by configuration in Airflow.
• Worked to create a data validation and integration testing framework using docker.

Feb 2019 – Mar 2020 · 1 yr 1 mo

• Built U-net based Generator and a patch-gan Discriminator to remove clouds from Sentinel-2 images by fusing Sentinel-1 SAR data.
• Added artificial clouds by simulating Perlin noise
• Tools Used: Keras, Tensorflow, Scikit Image, Numpy, Noise, Jupyter Lab,

Dec 2018 – Mar 2020 · 1 yr 3 mos

• Build a framework for deriving farmer credit score using Demographic, Transactional and Satellite data.
• Tools Used: Numpy, Pandas, XGBoost, Logistic Regression, Python, visualization libs(Seaborn, Matplotlib), Excel.

Sep 2018 – Mar 2020 · 1 yr 6 mos

• Built FarmRating aggregating different factors such as cropping pattern, farm stats, connectivity stats, soil stats etc
• Providing flexibility to incorporate different weight depending on the preference of various industries like banks, seeds, the fertilizer industry etc.
• Tools Used: Multivariate Regression, R (rgdal, raster, rgeos), PCA, t-SNE

Jun 2018 – Mar 2020 · 1 yr 9 mos

• Postgresql Data insertion to support geographical queries by building spatial indexes
• Tools Used: Python, Postgres, PostGIS, Redis, S3, apache-airflow

May 2018 – Mar 2020 · 1 yr 10 mos

• Trained model on Nvidia GPUs and monitoring progress on Tensorboard
• Processing of raw edge maps to produce clean edge maps using image processing techniques
• Accuracy improvement by using historical imagery to produce strong and weak edges
• Tools Used: Deep Learning (CNN, Keras-HED) Python, R, Skimage, OpenCV, GIS (GDAL/OGR), Database (Postgres), Image Source (Google Map Images), Tensor Board, Docker

Jan 2018 – Mar 2020 · 2 yrs 2 mos

• Sourced RGB Images from google earth for 18 states for the past 10 years
• Scaled the system across 1500 servers to increase downloading speed to 90 images per sec using python GUI controller libraries
• Tools Used: Google Earth Engine, pyautogui, Redis, AWS SQS, MySQL, GDAL, OpenCV

Sep 2017 – Mar 2020 · 2 yrs 6 mos

• Edge maps to identify different farms in high-resolution images
• Removal of waste segments using shape filtering by HU moments
• Employing historical imagery to identify land fragmentation
• Tools Used: Python, Numpy, Sci-kit learn, pandas, k-means clustering on shapes, multiprocessing

Jun 2017 – Mar 2020 · 2 yrs 9 mos

• Built OCR to read metadata in satellite imagery with an accuracy of 95% and processing speed of 5 images per sec
• Tools Used: Python, Scikit-learn, OpenCV, matplotlib, k-means clustering, SVM, pyautogui

Jul 2016 – May 2017 · 10 mos · Gurugram, Haryana, India

• PUSH NOTIFICATION SYSTEM
• Project description- Setting up a bridge between different mosquito message brokers to create a highly available system in the event of a node failure.
• Handled Topic explosion problem, to implement push notification system to a dynamic user segment.
• Built a stress testing environment to simulate real-life scenario for mosquito broker based on MQTT, connecting 50K devices.
• Tools Used: MQTT message broker, Java, Bloom Filter, Vagrant, HBase, Hashing
• DATA MODELLING
• Probabilistic Data Structure buffers computation supporting intersection across different user features and union across different timepoints to produce aggregated result.
• Exploited FP growth to capture user segments that account for maximum traffic on the client application.
• Facilitated the availability of a Custom White List to engage with users segments, which are of particular interest to the app owners.
• Storing data at different granular levels to reduce buffers needed for result aggregation.
• Tools Used: SPARK, Scala, Cloudera, HBase, FP Growth, Bloom Filter, Minhash, HyperLogLog, Count-Min Sketch, SBT, Jaccard Similarity Index, UDAF
• QUERY LAYER
• Used MemCached and Guava Cache to implement two-level caching layer for attaining near real-time response
• Buffer selection methodology to minimize the no of buffers on which processing needs to be done
• WEB SCRAPPING
• Scrapped Maven Coordinates corresponding to multiple jars from Maven Repository for migrating an ant-based project.

May 2015 – Jul 2015 · 2 mos · Gurugram, Haryana, India

• Studied Atmospheric and Vacuum Distillation Column installed in Kochi Refinery Expansion Project
• Simulated the Recovery of Sulphur containing compounds in the Tail Gas Treatment Unit using Aspen
• Detailed analysis of pump hydraulics and subsequent pump type selection for BPCL Bombay