Location via proxy:   
[Report a bug]   [Manage cookies]                

Akshat Surolia

Home

Skills

Experiences

Education

Projects

Contact

Akshat Surolia

Akshat Surolia

AI Researcher  |  

India

Who Am I?

Experienced Machine Learning Engineer and AI Researcher with 3 years of expertise in developing and deploying cutting-edge AI solutions across a variety of domains including education, marketing, medical, manufacturing, mental health, IT and more. Skilled in leveraging different modalities including text, speech, and images to create innovative AI models.

Experience

  • June, 2022 - Present

    WysaMachine Learning EngineerBangalore, india
    • Elevated performance of pre-existing models through targeted enhancements.
    • Pioneered a novel architecture for intent classification utilizing semantic similarity obtaining in 300% improvement in classification accuracy.
    • Expertly trained and fine-tuned models to refine aptness performance.
    • Utilized advanced techniques such as quantization, graph transformation, and node fusion to drive model optimization resulting in a 10x reduction in server cost.
    • Contributed to the development of patents and internal research projects.
    • Provided support for ongoing multilingual projects.
    • Conducted extensive user behavior analysis to drive retention and engagement.
    Python
    PyTorch
    Flask
    FastAPI
    Docker
    AWS
    Mongo DB
    Natural Language Processing
    Text Summarization
    Zero Shot Classification
    Sentiment Analysis
    User Embeddings
    Speech Synthesis
    Language Translation
    Semantic Similarity
    MLOps
    System Design
    Generative AI
    LLMs
    PEFT
    Quantization
    User Profiling
    Data Analysis
    AI Explainability

  • August, 2021 - June, 2022

    DSMATICSData ScientistPune, india
    • Building DL models and pipelines for use cases across facets spanning e-commerce, consumer data, markets, education, medical, industrial manufacturing etc.
    • Scrutinizing existing Machine Learning (ML) models to identify key areas of modifications and deliver optimal solutions.
    • Developing and implementing Machine learning operations (MLOps) framework to streamline the process of model development, deployment, and optimization.
    • Creating synthetic datasets as part of developing models and slashing data acquisition costs by 95%.
    • Consulting Companies for AI integration to meet their needs and resolve problems by recommending data-driven solutions.
    • Training cooperate employees in the domain of data science to meet the current market needs.
    Python
    Tensorflow
    PyTorch
    Flask
    FastAPI
    Natural Language Processing
    Computer Vision
    Text Summarization
    Zero Shot Classification
    Clinical Data Classification
    Question Generation
    Sentiment Analysis
    Image to Image Translation
    Image Generation
    Object Detection
    Semantic and Instance Segmentation
    Latent Space Disentanglement
    Image Matting
    Image Processing
    User Embeddings
    Content Based Filtering
    Speech Synthesis
    Speech Recognition
    Recommendation System

  • Janauary, 2021 - June, 2021

    Necesario InnovationsComputer Vision EngineerIIT Gandhinagar, India
    • Implemented real-time HDR on Jetson Nano and applied Contrast Enhancement.
    • Developed and Optimized models for 3D model Reconstruction, 3D Photography, Super Resolution, and Image Enhancement by applying 8-Bit Quantization, and using EfficientNetV2 feature maps to increase the model efficiency by 40% and making the inference 5 times faster.
    • Lead Programmer and Technical Head of the product ‘Snapper’.
    Python
    Computer Vision
    TensorFlow
    PyTorch
    Flask
    OpenCV
    GStreamer
    Jestson Nano
    HDR Capturing
    Image Enhancement
    3D Image Modelling
    3D Image Photography
    Depth Estimation
    Neural Filters
    Image Processing
    Image Segmentation

  • Janauary, 2021 - June, 2021

    TechnocolabsDeep Learning EngineerIndore, India
    • Led the team of 11 members to develop a complete chat-bot with Dialogflow aiding the users with their tech queries.
    • Supervised the team, designed the pipeline and developed a webhook to construct a query-based extractive summarization probabilistic model, achieving 94.2% accuracy.
    • Assigned as a Team Lead.
    Python
    Dialogflow
    TensorFlow
    Flask
    Natural Language Processing
    Webscraping
    Time Series Prediction

    • Projects

    • Clinical ICD-10 Code PredictionFebruary, 2022 - February, 2022

      Tech Stack

      Python
      Clinical BERT
      ICD-10
      Fast API
      Hugging Face
      Repository

      Project Summary

      • This repository includes API to get ICD-10 codes from descriptions. The Model is a transformer based on BERT. The embeddings are being finetuned based on the following paper: https://arxiv.org/abs/1904.03323.
      • Download the finetuned model from Hugging Face: https://huggingface.co/AkshatSurolia/ICD-10-Code-Prediction.
    • FaceMask Classification with SOTA ModelsFebruary, 2022 - February, 2022

      Tech Stack

      Python
      Vision Transformers
      PyTorch
      Fast API
      Repository

      Project Summary

      • This repository holds an API made with FASTAPI that delivers the downstream task of Face Mask Classification performed on Self Currated Custom Dataset with various State of the Art deep learning models like ViT, BeIT, DeIT, ConvNeXt, VGG16, EfficientNetV2, RegNett and MobileNetV3.
      • Download the finetuned model from Hugging Face: https://huggingface.co/AkshatSurolia.
    • Real-Time Call Transcription and Sentimental AnalysisJanauary, 2022 - February, 2022

      Tech Stack

      Python
      Twilio
      Media Streaming
      Wav2Vec2.0
      BERT
      Speech To Text
      Sentiment Analysis
      Repository

      Project Summary

      • Real Time Call Transcription and Sentimental Analysis, Call Initiates with Twilio, Streams in MuLAW, Preprocessed and converted into .wav, Transcribed with Wav2Vec2.0 and Analysing sentiment with BERT.
      • Implemented in Python and PyTorch framework and Flask websocket.
    • Diverse Facial EditSeptember, 2021 - November, 2021

      Tech Stack

      Python
      Stylegan
      Stylegan2
      Styleclip
      Latent Space Disentaglement
      Encoding
      Face Editing
      Repository

      Project Summary

      • Diverse Facial Edit with StyleGAN, StyleGAN2, StyleClip with ViT, and Other Features like Background Removal and Face Swap.
      • This API serve the purpose of generating random real looking images of human faces which can be edited in latent space and can generate new images from the given parameters with extra feauture of removing background and faceswap.
    • 3D-Photography with Image Inpainting and Depth EstimationMarch, 2021 - March, 2021

      Tech Stack

      Python
      PyTorch
      OpenCV
      Image Inpainting
      Depth Estimation
      Repository

      Project Summary

      • Comparing depth estimation by Tiefenrausch and PyDNet respectively. Sensing Depth from 2D Images with Tiefenrausch Network.
      • Inpainting Background behind the Foreground objects to create 3D Photos with Parallax Animation.
    • Image Enhancement and Super ResolutionJanauary, 2021 - February, 2021

      Tech Stack

      Python
      PyTorch
      OpenCV
      Image Enhancement
      Super Resolution
      Repository

      Project Summary

      • Using GANs to learn the translation function using a residual convolutional neural network that improves both colour rendition and image sharpness, ultimately achieving Image Enhancement with Contrast, Illumination, Saturation Adjustment and Noise Reduction.
      • Interpolating images to upscale resolution via Super Resolution Models such as EDSR, FSRCNN, SRGAN, and CAR
    • Ooze - Handwritten Text GeneratorJuly, 2020 - December, 2020

      Tech Stack

      Python
      TensorFlow
      OpenCV
      Cyclegan
      Handwritten Text Generation
      Instance Normalization
      Repository

      Project Summary

      • This is an TensorFlow Implementation of Ooze - Handwritten Text Generator by Akshat Surolia, This is an Image to Image translation approach to generate realistic hand written sentences by taking text as an input.
      • Implemented in python and TensorFlow framework, deployed on Heroku.
    • Ace TechSeptember, 2020 - October, 2020

      Tech Stack

      Python
      Dialogflow
      Chrome Extension
      Webscraping
      Information Extraction
      Summarization
      Repository

      Project Summary

      • AceTech is a chat bot used as web extension to serve tech and programming queries with concise answers up to 100 words and video/articles cards.
      • Implemented in Python and Dialogflow with real time web scraping and applying NLP techniques for query based extractive summarization, deployed in Heroku using Flask API
    • Digitalizing Prescription ImageMarch, 2020 - May, 2020

      Tech Stack

      Python
      TensorFlow
      OpenCV
      OCR
      Image Processing
      Repository

      Project Summary

      • Digital transformation of hand-written prescription text using advance image processing techniques and deep learning methods.
      • The neural network model consists of, convolutional neural network for feature extraction, recurrent neural networks for dealing with character’s sequencing. We use connectionist temporal classification loss function.
    • Maze RunnerSeptember, 2019 - November, 2019

      Tech Stack

      Python
      Dialogflow
      Tkinter
      Maze Solver
      Repository

      Project Summary

      • Maze Runner is based on Travelling Salesman Problem, to minimize the travelling cost to find the best possible solution. The user input maze image of m x n dimension which is converted to a graph, extracting the nodes and making connections which is used to traverse the maze via different algorithms (A-star, Dijkstra BFS, DFS).
      • Compare different algorithms to get the most efficient path, while generating nodes in a graph. The maze can be a source to find the shortest easy path, from the multiple path that leads to the solution.
    • EdversitySeptember, 2019 - November, 2019

      Tech Stack

      Java
      TensorFlow
      Keras
      Machine Learning
      Mordern UI
      Python
      Repository

      Project Summary

      • Edversity is a learning application based on android platform. featuring over 9 modules supporting visual and haptic feedback and a responsive UI.
      • Implemented in Java, Python, Firebase, TensorFlow and Google API.

    Skills

    C is a general-purpose, procedural computer programming language supporting structured programming, lexical variable scope, and recursion, with a static type system.

    Cascading Style Sheets (CSS) is a style sheet language used for describing the presentation of a document written in a markup language such as HTML.

    HTML5 is a markup language used for structuring and presenting content on the World Wide Web.

    Java is a high-level, class-based, object-oriented programming language that is designed to have as few implementation dependencies as possible.

    JavaScript often abbreviated JS, is a programming language that is one of the core technologies of the World Wide Web, alongside HTML and CSS.

    Python is an interpreted high-level general-purpose programming language.

    SQL (Structured Query Language) is a domain-specific language used in programming and designed for managing data held in a relational database management system (RDBMS).

    Dialogflow is a natural language understanding platform used to design and integrate a conversational user interface into apps.

    FastAPI is a Web framework for developing RESTful APIs in Python.

    Firebase is a platform developed by Google for creating mobile and web applications.

    Flask is a micro web framework written in Python. It is classified as a microframework because it does not require particular tools or libraries.

    GitHub, Inc. is a provider of Internet hosting for software development and version control using Git.

    OpenCV is a library of programming functions mainly aimed at real-time computer vision. Originally developed by Intel.

    PyTorch is an open source machine learning library based on the Torch library.

    Hugging Face is an open-source framework for natural language processing (NLP) that provides state-of-the-art models and tools.

    React is a free and open-source front-end JavaScript library for building user interfaces based on UI components.

    TensorFlow is a free and open-source software library for machine learning and artificial intelligence.

    Docker is a platform that allows you to package and distribute applications in isolated containers for efficient and consistent deployment across different environments.

    Kubernetes is an open-source container orchestration platform that automates the deployment, scaling, and management of containerized applications, providing a highly scalable and resilient infrastructure.

    Redis is an open-source in-memory data structure project implementing a distributed, in-memory key-value database with optional durability.

    Amazon Web Services, Inc. (AWS) is a subsidiary of Amazon providing on-demand cloud computing platforms and APIs to individuals, companies, and governments, on a metered pay-as-you-go basis.

    Google Cloud Platform (GCP), offered by Google, is a suite of cloud computing services that runs on the same infrastructure that Google uses internally for its end-user products.

    Heroku is a cloud platform as a service (PaaS) supporting several programming languages.

    IBM cloud computing is a set of cloud computing services for business offered by the information technology company IBM.

    MongoDB is a source-available cross-platform document-oriented database program.

    PostgreSQL is a free and open-source relational database management system (RDBMS).

    Visual Studio Code is a source-code editor made by Microsoft for Windows, Linux and macOS.

    Linux is a family of open-source Unix-like operating systems based on the Linux kernel.

    Education

  • June, 2019 - June, 2021

    Gujarat University
    Department of Computer Science
    Department of Computer Science
    M.Sc. in Artificial Intelligence & Machine LearningAhmedabad, India8.3 CGPA
    • Volunteered in Tech Kaushalya, Tech Fest of Gujarat University.
    • Mentor at SWoC 2021, Open Source Contribution

  • June, 2016 - May, 2019

    Faculty of Computer Applications and Information Technology
    GLS University
    Faculty of Computer Applications and Information Technology
    Bachelors in Computer ApplicationsAhmedabad, India9.07 CGPA
    • Awarded Best Student of Batch by GLS University.
    • Attained 1st position in Smart Gujarat Hackathon in Health and Family Welfare Department.
    • Won Relay Programming (Java), in a National Level Tech-Fest event.
    • Secured 1st position in Extempore, in a National Level Tech-Fest event.

  • June, 2015 - May, 2016

    CBSE
    St. Joseph’s Convent Sr. Sec. School
    St. Joseph’s Convent Sr. Sec. School
    Higher Secondary Certificate – 12thRatlam, India9.4 CGPA
    • Awarded with the title of Mr. Talented.

    • Publications

    GLS KALP – Journal of Multidisciplinary Studies
    Ooze - Handwritten Text Generator
    29 December,  2021
    How to cite

    See Publication

    Surolia, A. (2021). Ooze - Handwritten Text Generator. GLS KALP – Journal of Multidisciplinary Studies, 1(4), 35–49. Retrieved from https://glskalp.in/index.php/glskalp/article/view/19

    In this paper we show how Generative Adversarial Network (Goodfellow et al., 2014), more specifically Cycle-GAN(Zhu et al., 2017), can be used for Human like handwriting generation with output size up to a text line. The methodology used in Cycle-GAN is to establish translation between two different domains (e.g., connection between image of horse and zebra), extending this methodology to establishment of translation between machine printed text and handwritten text is mentioned in this paper. The neural network used in this paper is trained for dataset created by the author and can be trained with another dataset.
    GLS KALP – Journal of Multidisciplinary Studies
    Recommendation System based on Artist and Music Embeddings
    01 July,  2022
    How to cite

    See Publication

    Surolia, A. (2022). Recommendation System based on Artist and Music Embeddings. GLS KALP – Journal of Multidisciplinary Studies, 2(3), 8–15. Retrieved from https://glskalp.in/index.php/glskalp/article/view/33

    In this paper, we present a personalized music recommendation system based on the embeddings of artists and music. The main peculiarity of our work is that the determining factors of a user’s preferences for a developed music recommender system are the artists and the music that they listen to. The artist embeddings inform the network about the contextual representation of artists in a latent space, where similar artists are closer to each other. The music embeddings hold the information about the music. Both embeddings are then combined to form a new embedding, which is then used to predict the user’s preferences. We use the Spotify’s API to collect the data to train and evaluate the model. Two approaches of building a music recommender system are considered in this paper. Each approach significantly differs in the way the embeddings are learned.
    Get in Touch