Rohan Dawkhar

Rohan Dawkhar

About

Detail

Technical Writer at Medium
Maryland, United States

Timeline

work
Job
school
Education

Résumé

Jobs verified_user 0% verified
  • Medium
    Technical Writer
    Medium
    Jun 2023 - Current (3 years 5 months)
  • Unify
    Open Source Contributor
    Unify
    May 2023 - Oct 2023 (6 months)
    ◦ Paddle Frontend: Add math, tensor, activation, linear algebra functions to the PaddlePaddle frontend. Write the corresponding test cases for the same and test the functions on varied input samples. Push and create a pull request for the code snippets after successful test cases. ◦ Community: Interact with the Unify team and community on their Discord channel to brainstorm ideas, help to solve beginner friendly issues for newcomers, etc. ◦ Blogs: Write blogs on medium to contribute to the betterment of Unify’s open source community. Explaining how to setup repository on codespaces, tackling excessive RAM consumption while using Docker, etc.
  • Kampd
    Data Science Intern
    Kampd
    Jun 2022 - Dec 2022 (7 months)
    ◦ Recommendation Engine: Build and tested utility endpoint feature for Content Abstraction Module (specifically for a delete feature) of Recommendation Engine in DEV Environment. ◦ Data Annotation: Did Data Annotation on acquired users’ data which was later used as a testing set for Taxonomy Engine Prototype. ◦ Taxonomy Engine: Build and tested Taxonomy Engine which appropriately labels the contents (based on Title, description and user Tags) as per the taxonomic structure and requirements with an inherent capability to learn from the incoming Data and predict the labels accordingly as per the changing distributions which would solve the need of updating the label set of the Taxonomy engine manually. This Engine was designed with an ideolog
Education verified_user 0% verified
  • University of Maryland
    Master of Science - MS, Data Science
    University of Maryland
    Jan 2024 - Mar 2026 (2 years 3 months)
    Coursework: Fall 2025: DATA607 - Communication in Data Science and Analytics DATA641 - Natural Language Processing DATA650 - Cloud Computing Spring 2025: DATA604 - Data Representation and Modeling DATA605 - Big Data Systems DATA606 - Algorithms for Data Science Fall 2024: DATA601 - Probability and Statistics DATA602 - Principles of Data Science DATA603 - Principles of Machine Learning
  • A
    Certificate in Applied AI, Data Science, Machine Learning, Deep Learning
    Applied Roots
    Jan 2020 - Dec 2022 (3 years)
  • Savitribai Phule Pune University
    Bachelor's degree, Computer Engineering
    Savitribai Phule Pune University
    Jan 2016 - Dec 2020 (5 years)
  • G
    HSC(12th, Science Stream
    GRPS Junior College Pune
    Jan 2014 - Dec 2016 (3 years)
Projects (professional or personal) verified_user 0% verified
  • C
    Celebrities Face Clustering
    • Clustered cropped celebrity faces using 128-dimensional embeddings using a face recognition library. • Generated multiple montages for visual inspection and saved the cluster montages in different folders with IDs. • Technologies: Python, TensorFlow, Face Recognition Library, NumPy, Matplotlib, computer vision.
  • D
    Document Classification with CNN.
    Objective:- To classify all the 18,828 documents into any one of the 20 classes. Solution:- The dataset used is 20 Newsgroup Dataset which contains different text documents belonging to a particular class. I started the project by preprocessing the data, then tokenizing, sequencing, padding the preprocessed text and used predefined Glove Vectors. Using tensorflow, built the architecture of the model and trained the model and got 0.72 F1 Score(word embeddings).
  • M
    Microsoft Malware Detection using GCP.
    Objective:- To identify whether a given file/software is a malware. Solution:- Used Google Cloud Platform(GCP) for processing the data and training the model, as the dataset was approximately of size 200GB. Featurized both byte files(used SelectKBest to select top k bigrams features) and asm files(pixel intensity features). Modeled the data using XGBoost Classifier and got around 0.02 log loss on the test data.
  • F
    FoodLens - Interactive Digital Menu Generator
    • Vision LLM (Meta LLaMA via OpenRouter) extracts menu items, descriptions, and costs from input menu images and converts the information into a JSON format. • Unsplash API fetches high-quality dish images with the help of the menu item name. • Web interface displays an interactive digital menu with enhanced details. • Technologies: Python, FastAPI, Vision LLM, Unsplash API, HTML/CSS.
  • L
    LSTM on donors choose dataset.
    Objective:- To predict whether a DonorsChoose.org project proposal submitted by a teacher will be approved or not, using LSTM layer. Solution:- Vectorized the already preprocessed data(text data, categorical data, numerical data) and embedded the text data using predefined Glove Vectors. Defined various model architectures using tensorflow layers as well as used custom callbacks. Used LSTM(Long Short Term Memory) layer in the architecture. Obtained around 0.74 AUC Score for test data for all the models.