Mar 2023 – Present · 3 yrs 3 mos · Finland · Remote

• ● Developed an AI-based translation system encompassing 100+ languages for an actionable competitive market intelligence platform, resulting in a 2x performance improvement and 50% cost reduction, which is achieved by deploying a multilingual model on a single machine handling thousands of requests daily.
• ● Designed and deployed an end-to-end industry insights classification pipeline on an Azure Kubernetes cluster as a real-time online endpoint, achieving a 50% improvement in accuracy compared to the previous classifier.
• ● Integrated a benchmarking pipeline to continuously enhance model performance in a production environment.

Aug 2022 – Mar 2023 · 7 mos · Texas, United States · Remote

• ● Contributed to developing an AI-based sales intelligence platform that enhanced Sales Performance Management (SPM) for startups and companies, leading to more efficient sales calls and doubling client conversions in less time.
• ● Designed and implemented end-to-end generic Kinesis routing pipelines, including unit testing and detailed documentation.
• ● Build, train and deployed ML pipelines for Masked Language modeling and classification on sales data using various transformer architectures, resulting in understanding of sales context in detail and accuracy up to 95%.
• ● Engineered real-time speech-to-text and speaker diarization capabilities for sales meetings.
• ● Developed cost-efficient, low-latency lambda containers for multiple microservices, optimizing high-level API code.

Oct 2021 – Aug 2022 · 10 mos · Gurugram, Haryana, India · Remote

• ● Worked on legal text analytics project, extracting 10+ key clauses from legal agreements such as MSA, PSA, Nodal, licenses, and more.
• ● Implemented OCR solutions using Tesseract for digitizing and processing scanned legal documents, enabling automated text extraction from image-based contracts and agreements to support downstream NLP pipelines.
• ● Built and fine-tuned BERT-based pre-trained NER models to detect custom entities like person names,
• organizations, and addresses.
• ● Deployed an end-to-end model on AWS Lambda using Docker while maintaining an Elasticsearch database.

Mar 2021 – Jun 2021 · 3 mos · Bratislava, Slovakia · Remote

• ● Consulted with users, vendors and technicians to determine computing
• needs and system requirements.
• ● Analyzed problems to develop solutions involvingcomputer hardware andsoftware.
• ● Developed Real time violence detection system using
• Tensorfow/Keras andNvidia Tensorrt on RWF-2000 dataset.
• ● RWF-2000 is a database with 2000 videos captured by surveillance
• cameras in real-world scenes.

Jan 2021 – Mar 2021 · 2 mos · Kyoto, Japan · Remote

• ● Conducted comprehensive research, synthesized information from multiple sources, and presented actionable results.
• ● Collaborated closely with an Assistant Researcher on her thesis, maintaining optimal communication to effectively and efficiently complete project.
• ● Researched and implemented best practices for a Character-Based Neural Machine Translation System to translate from Minangkabau to Indonesian using TensorFlow, achieving over 95% accuracy. Leveraged the close phonetic similarities but distinct spelling differences between the two languages to enhance translation performance.
• ● Experimented with Char-Transformers, Char-RNN, BiLSTM, Sentencepiece, and Byte Pair Encoding, ultimately achieving the best results with Char-Transformers on a 14,000-entry word-to-word dictionary dataset.

Sep 2019 – Oct 2020 · 1 yr 1 mo · Delhi, India

• ● Analyzed Twitter data using statistical and machine learning techniques to provide insights into trending topics in India and applied robust topic modeling with Top2Vec on COVID-19 tweets and visualized the results.
• ● Experimented with various machine learning and deep learning models for sentiment analysis, including pQRNN (Projection-based Quasi-Recurrent Neural Network), achieving over 92% accuracy on the Jigsaw Toxic Comment Classification dataset with a model size of just 320KB.
• ● Developed an aspect-based sentiment analysis model using LCF-BERT, incorporating local context embeddings to determine aspect sentiment and deployed it end-to-end using serverless AWS Lambda and API Gateway.