tushar kamble kamble

• Spearheaded a machine learning team of 4 to engineer an automated recruitment system, crafting core algorithm and implementing Power BI for impactful visualization, enhanced candidate selection processes and boosted recruitment efficiency.
• Guided a team of 4 developers to improve feature selection from over 30, 000+ data points executing EDA and Select K Best, optimized with hyper-parameter tuning.
• Trained diverse ML models (KNN, k-means, SVM, Random Forest) across varied data types, incorporating LDA, NER, Spacy, TensorFlow and BERT to elevate document classification and recommendation systems.
• Employed neural networks (YOLO v5, R-CNN, SSD) for video and image transforming, enhancing object detection accuracy by 60% and streamlining interfaces via Streamlit.
• Constructed complex chatbot systems employing large language models (GPT-3.5 turbo, GPT-4, Llama 2), tailored smaller models (BERT, DistilBERT) for specific tasks, combining Azure QnA, and optimizing production workflows.
• Implemented a planned deep learning speech-to-text model with sentiment analysis, elevating chatbot responsiveness and emotional acuity, scaling to process 10, 000+ hours of audio data.
• Deployed Machine Learning models and collaborated with 3 technical leads on AWS, utilizing Lambda, S3, MLflow, and SageMaker integrated with Docker, expediting production pipeline integration.
• Fine-tuned various LLM's, BERT, Llama, Mistral models applying LORA, QLORA and IA3 technique via LangChain and Llama Index, improving model performance by 40% with Rouge and BLEU scores.
• Built CI/CD pipelines with MLflow for deploying RAG architecture, ensuring seamless amalgamation and real-time responses; and trained 6 interns in deployment protocols.
• Amplified automation using Docker, Flask, OpenAI, Python on Azure Databricks, combining Kubernetes, Spark, PostgreSQL, reducing app development and deployment from 4 days to 2 mins.
KEY PROJECTS
• Advanced Document-Handling Chatbot: Designed and managed creation of an NLP chatbot for document processing, leveraging NLTK,
SpaCy, and scikit-learn, and integrating AI models (GPT-3.5 Turbo, Llama 2, BERT) using FAISS to handle 1M+ documents on Azure.
• NASCAR Pitstop Time Optimization: Collaborated with 4 business managers to applying neural network models (YOLO v5, R-CNN, SSD),
enriching operational efficiency by 25 % and performance analysis across multiple metrics.
• Loan Default Risk Prediction: Developed Python-based predictive models (logistic regression, random forests, XGBoost, AdaBoost, LSTM)
for credit risk, administering user requirements into dynamic PowerBI dashboards, maximizing functionality by 45%.
• Pneumonia Detection with Deep Learning: Refined and delivered pneumonia detection system using TensorFlow/CNNs; achieved 95%
accuracy, streamlined user interface in Streamlit for seamless diagnostic experience.
• Speech-to-Text Chatbot: Analyzed and executed speech-to-text system on Azure DevOps, reduced latency by 30%, scaling to handle 100K
daily queries, incorporating NLP technologies and deep learning architectures.
• Knowledge Graph-Based Query Extraction Chatbot: Mentored 7 junior data scientists to architect knowledge graph-based query extraction
chatbot using Neo4j, LLMTransformer and GPT-3.5 via LangChain and Llama Index.