Smritika Sadhukhan

Equinix Public Sector -

• Performed topic extraction and fine tuned topic representation for Deltek opportunity (document with business requirements) and equinix corpus using semi-supervised model BERTopic
• Improved the existing ontology (Equinix relevant topic and respective keywords)
• Filtered out Relevant documents with 86% F1-score on the trained deltek corpus (100k) using SetFit with only 50 samples.
• Skills - Bert , BERTopic, SetFit, keyBert, KeyVectorizor.

Healthcare chatbot -

• Deployed a Chabot created using Dialogflow API to ask intelligent questions and predict the disease based on symptoms and help user to book appointment with doctors

Recommendation system -

• A recommendation system on Amazon products using unsupervised learning and similarity matrix and got similar results when compared with Amazon
• Skills – Python, SOM, Gaussian mixture, Cosine similarity, flask, heroku

Voice Assistant -

• A SAAS where any non technical person can create data in CSV format and leverage the benefit of dialogflow and Rasa by single click without even using them and get a voice assistant instantly and used open source models, Google recognition,etc for Speech to text and text to speech conversion
• Skills – Python, Heroku, Flask ,Rasa, Dialogflow , Docker, STT, TTS

Face recognition -

• Performed image augmentation. Using dlib face recognition model extracted the face embedding and trained on machine learning model. Successfully handled twin classification problem. Got 95% accuracy on current employee of our organization
• Skills – Python, Heroku, Face recognition, facemesh

Effortless Travel buddy and a SAAS model for a voice-bot Built

• Advanced multilingual Natural Language Processing and Natural Entity Recognition model for multi-purpose
• Skills -Single handedly created script for Data creation, Data processing, LSTM, Bidirectional-LSTM,CNN,CRF, Embedding, Packaging, flask, tensorflow serving, Docker, API ,postman, MURIL, YOLO5

Recommendation system

• Building various recommendation system like friends, followers and post using content and collaborative method or Social Media Platform Friend App

Depression Detector

• Used Acoustic features of various individuals and calculated the score of suffering of the individual with depression

Prediction of alloy ratio

• Predicted accurate alloy ratio using machine learning boosting algorithm and deep learning to reduce manual human effort and further increase engine performance and its durability