IQVIA, a Fortune 500 company, drives innovation in life sciences with advanced analytics, technology solutions, and clinical research, employing 86,000 people across 100+ countries.

The projects I was involved in: Shallow Parsing and Deep Parsing: - Developed data preparation, training pipelines, alongside a joint modeling pipeline.- Prepared inference pipeline with PyPI package, API (using FastAPI), and docker file.- On Penn Treebank (English) achieved SOTA results for both shallow (96.47 FScore) and deep parsing (UAS: 97.49%, LAS: 96.36%).Human Monitoring (Computer Vision):-Developed live system for monitoring people from office lobby camera.-Utilized YOLOv8 for human detection, DeepSORT for tracking, and ArcFace with cosine similarity for face recognition.

The projects I was involved in: Sentiment Analysis:-Conducted extensive model experimentation, cross-dataset analysis, statistical significance tests, etc.-First-authored the paper "SentiGOLD: A Large Bangla Gold Standard Multi-Domain Sentiment Analysis Dataset and its Evaluation" published in KDD '23.Named Entity Recognition: -Developed the baseline model using BERT, BiLSTM, CRF.-Experimented with different class imbalance handling techniques such as resampling, curricular-face loss etc.-Developed a Bangla gazetteer (lists of place names, people's names, and other named entities) and combined with baseline.Lemmatization:-Collaborated with linguists to develop a word formation dependent rule-based lemmatizer for Bangla (submitted in EMNLP '23).-Achieved 98.17% accuracy on internally built test dataset.Language Model:-Developed pre-training boilerplate for BERT, XLNet, and Word2Vec using hugging face libraries.Banglish Classifier:-Implemented a CNN model to classify text as romanized (Banglish) or not and achieved 0.93 macro F1 on internal dataset.