*** * *********** **, *** C*

Bloomington, IN ****8

https://www.linkedin.com/in/harsha-

renkila-038539bb/

HARSHA RENKILA

812-***-****

advpjj@r.postjobfree.com

EDUCATION

Indiana University Bloomington Bloomington, In Jan 2022 – August 2023 (Expected)

• M.S. in Computer Science, GPA: 3.95/4

• Teaching Assistant for Applied Algorithms Course University Of Hyderabad Hyderabad, India July 2014 – Aug 2019

• Integrated Master of Technology in Computer Science, CGPA: 8.24/10 Relevant Coursework: Design and Analysis of Algorithms, Computer Vision, Machine Learning BioInformatics, Pattern Recognition, Probability Statistics, Statistical Methods in AI, Image Processing, Text Processing, Data Mining, Advance Natural Language Processing, Neuro Engineering.

Honors: Gold Medal for Academic Excellence; Employee of the year(CRISP Award) WORK EXPERIENCE

Machine Learning Engineer-1 PhenomPeople PVT LTD, India Nov 2019 – Jan 2022

• Investigated and experimented various multi-lingual language models(Xlmr bert, multilungual bert, fasttext etc) to ex- tract(parsing) skills, experience, job type, etc using deep learning and NLP techniques from job description/resume. Deployed job parser which supports 108 languages, but went live for Spanish, Dutch, German, and French. Multi-lingual parser helped to acquire 10 new clients.

• Tracked and fixed bugs, re-factored the code, and redesigned the architecture for increased job parsing efficiency from 5 jobs/sec to 20 jobs/sec.

• Deployed and developed job-parser from scratch. SOTA nlp models like BERT, XLMR-BERT, MLP models, Spacy-Ner, Flair-ner, Fasttext, Flashtext, etc are used for text classification, NER, prediction, and text summarization tasks.

• Designed A/B testing mechanism to evaluate different recommendation widgets using custom NDCG metrics. Testing helped to improve user interaction with widgets.

• Experience in working with Git, Docker, deployment using jenkins, Spinnakar for kubernetes, and developed ml models using MLOPS pipeline.

• Implemented a framework to find bias like gender bias, ethnicity bias, Age bias, and skill Bias in the products. Deep learningmodelsaredevelopedtopredictgenderfromnameandethnicityfromnamewhichhelpedtoretaintheexisting clients.

• Recruited, mentored, and supervised a team of 3 to build end-to-end ML pipelines and integrate them into the Talent Recruiter platform.

Final Project University of Hyderabad, India July 2018 – June 2019

• Designed and implemented a novel framework for attention-based image generation for Telugu OCR- handwritten char- acters with Generative Adversarial Nets.

• Modeled a hybrid of Generative Adversarial Nets and Variational Auto-Encoders for image generation. Model showed a 9% increase in accuracy over Deep Recurrent Attentive Writer(DRAW).

• A research paper DRAW-GAN: An Adversarial procedure for sequential image generation and accepted by the IEEE TEN- CON 2019(International Conference).

Summer Intern IIIT Hyderabad, India May 2018 – July 2018

• Programmed python scripts to split compound sentences into simple sentences using Hindi language rules.

• Anusaaraka - A Neural Machine translator which is a central government-funded project that uses written scripts to translate to Dravidian languages.

Summer Intern IDRBT, India May 2017 – July 2017

• Attemped to compare traditional client-server architecture build using NodeJS with permission Blockchain architec- ture in Hyperledger fabric(an open source distributed ledger) on the bank transactions helps to find the challenges in migrating from the existing architecture.

ACADEMIC PROJECTS

• Multi-lingual Embedding Efficiency on ONET Data: Explored and interpreted the efficiency of muli-language embed- ding language models on domain specific Onet data using xlrm-Bert, MBERt, MUSE etc. Project details

• EmpathNet - Depression detection in semi clinical interviews: Designed and implemented siamese one shot learn- ing using different CNN networks like VGG16, Resnet, exception, etc. A combination of text features(BERT) and video features Empathnet model showed in par results with SOTA AVEC-2016 competition.

• Cancer Drug Response prediction: Proposed a new Restricted Boltzmann Machine(RBM) based visible neural network model to predict the response of drugs to cancer cells. Model turns out to increase Pearson correlation to 15% than vanilla vnn model.

• OMR Grading Assistant: Designed a tool that extracts answers and grades from marked OMR sheets using edge detec- tion, hough transforms, etc. Tool achieved 98% accuracy on grading.

• Anomaly detection in HDFS logs dataset: Developed an RNN-based model to analyze the HDFS logs to detect the anomaly and deployed it in real-time infrastructure for Quadratics PVT LTD.

• Dominant color detection in the catalog Images: Simple K-NN algorithm is developed to cluster and recommend similar catalog images. Achieved a 75% accuracy in recommendation.

• Helmet detection in the working labor: Annotated and implemented a model(using YOLO object detection model), that detects a helmet on the person and cautions the site manager.

• University Venue Management System: Built a library management system in Java and Oracle DB with a focus on triggers and procedures which offered management of 3 types of resources – books, cameras and rooms. SKILLS

Languages : Python, C, Bash, Java, Prolog.

Deep Learning Frameworks : Pytorch, Tensorflow.

Databases : MongoDB, SQL, ElasticSearch, Neo4j .

Web Development : HTML, CSS, Javascript, NodeJS.

LEADERSHIP

• Managed a teamof50tocreateanE-learning platform for Philosophy, funded by IndianGovernment(UGC). Coordinated production of E-learning content including lectures and assignments, and handled finances.

Contact this candidate