Karthik Revanuru

Data Science @ CRED.

Projects

• NLP modules for Noisy Data

  Built data extraction modules for various highly domain specific large contracts, that can even deal with noisy text (result of OCR). Built document categorization system that can automatically categorize and rename hundreds of different types of large documents and is extremely fast. Built a NLP module that can extract and cross verifies extremely sensitive information from multiple legal documents and reports the anomalies. Built a named entity recognition system using different machine learning and NLP techniques for highly domain specific data, which can work with very low amount of training data. Designed and built a novel algorithm using manchine-learning, that can extract structured data from hundreds of different types of tables like multi-line cells, cells separated only by a single space etc.

• Voice Assitant for Customer Support

  Built a voice assitant to answer customer queries in realtime. Have designed and built the entire pipeline from converting speech to text, classifing if it’s a pre existing query in database and final model deployment. Model was well received as it reduced work load on call center agents. One of the major challanges was sentence boundary detection and i solved it in an entirely differnt way and created an IP for it.

• NSFW detector

  Developed a NSFW detector which masks the NSFW content from video and images. I’ve done entire data annotation in an unsupervised way​ and created an IP for it

• Multi Modal Brain Tumor Segmentation

  Working on segmentation of intrinsically heterogeneous (in appearance, shape, and histology) brain tumors namely Gliomas from multi-institutional pre-operative MRI scans.Wrote python scripts to pre-process MR Images. Trained a Random forest to segment Gliomas with a dice coefficiant of 0.76. Developing a deep-learning based approach to generate better segmentations. Report

• Quora Question Pairs

  Thousands of new questions are posted in Quora everyday. In order to build a highquality knowledge base,it’s important to ensure each unique question exists on Quora only once. Writers shouldn’t have to write the same answer to multiple versions of the same question, and readers should be able to find a single canonical page with the question they’re looking for. In this project we applied Statistical Machine Learning and Deep Learning approaches to the problem of semantic matching on questions asked in Quora. Our model had an accuracy of 82 Percent. Report

• Speed Dating Analytics

  Performed Business Understanding, Descriptive Analytics, Cluster Analytics, Predictive Analytics using Classification and Association rules on a Speed Dating Experiment Dataset form Kaggle. Business Understanding, Descriptive Analytics, Association, Clustering, Classification.

• Automatic Shot Detection in Cricket

  We have detected four types of cricket shots namely Straight Drive, Cover Drive, Pull shot and Cut Shot in Cricket. Our HMM based model has an accuracy of 80 percent and we are currently working on rating these shots and give some feedback based on how good was the shot played.

• Harvest Prediction

  Successfully developed a predictive analytics system (Proof Of Concept) for Bosch Deep field Connect. Bosch Deep-field Connect assists the farmer in remotely monitoring the conditions on the field for strawberry and asparagus cultivation in real time and it also helps farmers in intelligent decision making with the motto of maximizing the harvest. I was responsible for predicting the harvest of the crop given soil conditions, temperature, moisture and harvest in the previous seasons.

• Electronic Health Record

  A digital version of a patient’s paper chart. This is real-time, patient-centered records that make information available instantly and securely to authorized users.

• Smart Neighborhood

  Built a social media platform centered around the idea to encapsulate the concept of a neighborhood where members in a neighborhood can convey and retrieve essential information efficiently. Built using Ruby on Rails.

• Heart Rate Detection using ECG Signal

  Wrote an efficient program in C++ and GNU Octave that takes raw ECG signal as input and detects and plots the R peaks from the QRS complex present in the signal using Fast Fourier Transform, High Pass Filtering and a Dynamic Sliding Window heuristic.

• Spectral Analysis of Accelerometer Data

  Use an Accelerometer to collect data during different activities such as jogging, walking, climbing stairs up/down, running ,riding a bus and provide the spectral analysis from accelerometer signal collected.

• Pedometer

  Built a Pedometer that uses gy521 sensor and Arduino. It counts the number of steps you take. It uses accelerometer and gyroscope to measure the steps.

• Locate Me- Disaster Relief App

  Two apps, one used by users which collects user location and the other used by authorities in case of a disaster where it will determine the last known location of victims.

• Driverless cars

  Developed a simulation for Driver-less cars which reflects real world traffic in Java.

• Virtual File System

  Implemented a basic Virtual File System to store and retrieve files and folders in C++

• Question Box

  Developed a chat system using TCP/IP protocol for asking questions to a teacher during lecture.

• Snail Mail Notifier

  Developed as a part of Basic electronics project a device which notifies when you get a Newspaper, Milk, Letters with different sounds.

• Adding New Instruction to Simple Scalar

  Added a new instruction to simple scalar PISA architecture and analyzed it’s performance on a set of benchmark programs.

• Digital Communication

  Simulated signal transfer over binary error erasure symmetric and AWGN channel of compressed (Discrete Cosine Transform, LZW, Huffman) and encoded (BCH/cyclic codes) music files (.wav).

• Operating Systems

  Implemented OS Scheduling Policies (FCFS, SJF, HRRN, Priority, RR, SRTN) in C. Created a shell in Ubuntu using C.

• Implementation of Common Algorithms

  Face detection, Matrix processing library, Huffman compression in Python, Sudoku solver using graph theory, Perfect hashing, Rabin-Karp string matching.