Extensively pursuing research opportunities to further enhance technical expertise and contribute to Samsung's intellectual property portfolio
Proficient in training and optimizing lightweight on-device models tailored for closed-domain question answering and text retrieval
Expanded Samsung Search's language support by implementing custom-trained models, enabling image-text search on edge devices
Enhanced and configured Large Language Model (LLM) inference speed to efficiently generate results for a substantial volume of prompts simultaneously while ensuring optimal latency
Authored a research paper on my work and submitted it to internal Samsung conference, while also filing a provisional patent application to safeguard the innovative use-case
Streamlined model integration by converting PyTorch codes to TensorFlow framework, enabling integration with available TensorFlow libraries.
Samsung Search
Software (Machine Learning) Engineer
01.2024 - 03.2024
Job overview
Dedicated to enhancing Samsung's on-device (offline) search by leveraging advanced algorithms and techniques that prioritize understanding the user's intent, as opposed to just the user's query
Developing a custom dataset that addresses domain-specific search challenges by collecting data from user manuals, blogs, and websites
Addressing data imbalance challenges by applying advanced sentence augmentation techniques and training a classification model using the triplet loss function
This approach has resulted in a model capable of accurately classifying triple-digit intents with an acceptable F1-score
Prioritizing technical efficiency, focused on optimizing the model to fit within the device's memory constraint of just 1 MB and achieve a search latency of under 20 ms
Recognized for exceptional performance and technical expertise, promoted to Senior Software Engineer within just 1 year and 2 months, bypassing the usual 2-year promotion cycle.
Intelligent Platforms and services
Summer Intern
05.2024 - 07.2024
Job overview
Identified and pre-processed an open-source dataset from Huffpost cataloging about 200K news headlines into 12 categories
Exploiting Pre-trained Glove Embeddings, trialed with different neural network architecture to build a multi-class topic classifier
Delivered a comprehensive comparative analysis of CNN, LSTM, Bi-LSTM, and multi-layer LSTM models, utilizing Tensorflow and Keras libraries, and optimizing evaluation metrics, such as AUC-ROC score, through the categorical cross-entropy loss function
Proposed attention mechanism to extract keywords from sentences to build a user-interest profile and personalized advertisements
Offered Pre-Placement Offer as Software Developer.
On-Device AI
Summer Intern
05.2024 - 07.2024
Job overview
Working on making a real-Time Water Health Monitoring Instrument version in an IoT based start-up focused on water resource management
Brainstormed with the core team and presented ideas to resolve difficulties coming in the final deployment of the product
Took the initiative to implement my solutions and worked on the design and manufacturing polypropylene caps capable of creating a separate lab-like environment for reading, calibration, and cleaning of sensors enabling their long term use in open environments
Recommended and tested ultrasonic cleaning mechanism to clean the sensor head with pre-set time intervals in real-time usage
Examined use of PGA460, 555-timer integrated circuits to customize ultrasonic generator circuit for transducers input.
Senior Software analyst - Machine Learning at Instituto de Pesquisas EldoradoSenior Software analyst - Machine Learning at Instituto de Pesquisas Eldorado