CODING & DEBUGGING SINCE — Y:2022
VASU
VASU
VASU
DESHMUKH
DESHMUKH
DESHMUKH
I am Vasu Deshmukh, an engineering student who enjoys exploring data, finding patterns, and building machine learning models using Python and SQL.
SCROLL
I’m interested in starting my career in data science. I enjoy turning raw data into insights, solving analytical problems, and building tools that help others make informed decisions.
I'm currently pursuing a Bachelor of Technology in Electronics and Computing Engineering at Mahindra University. Throughout my studies, I’ve focused on building practical skills in data analysis, machine learning, and statistical modeling using Python and SQL. My coursework and projects have helped me understand data-driven decision making, develop predictive models, and create effective visualizations.
Python, SQL, Java, C, Perl
The languages I use.
Python, SQL, Java, C, Perl
The languages I use.
Python, SQL, Java, C, Perl
The languages I use.
TensorFlow, Scikit-learn, Keras, NLTK, SpaCy, Transformers, Pandas, NumPy, Seaborn, Matplotlib
The libraries I use.
TensorFlow, Scikit-learn, Keras, NLTK, SpaCy, Transformers, Pandas, NumPy, Seaborn, Matplotlib
The libraries I use.
TensorFlow, Scikit-learn, Keras, NLTK, SpaCy, Transformers, Pandas, NumPy, Seaborn, Matplotlib
The libraries I use.
Jupyter Notebook, Google Colab, PyCharm, Git, GitHub, Docker, AWS, Power BI, Tableau
The tools and platforms I use.
Jupyter Notebook, Google Colab, PyCharm, Git, GitHub, Docker, AWS, Power BI, Tableau
The tools and platforms I use.
Jupyter Notebook, Google Colab, PyCharm, Git, GitHub, Docker, AWS, Power BI, Tableau
The tools and platforms I use.
PTHON
SQL
JAVA
C
PTHON
SQL
JAVA
C
Limurse
Remote
Jan 2025 — Mar 2025
Built a predictive content engagement model with scikit-learn and XGBoost to forecast optimal posting windows, improving engagement rates by 18% compared to baseline schedules. Developed an NLP-powered recommendation engine using transformer embeddings to cluster high-performing posts and suggest tailored content ideas, reducing creation time for marketing teams by 25%. Implemented an analytics dashboard and RESTful API (FastAPI + Plotly) delivering real-time campaign performance insights and AI-driven suggestions, enabling faster iteration cycles for clients.
Limurse
Remote
Jan 2025 — Mar 2025
Built a predictive content engagement model with scikit-learn and XGBoost to forecast optimal posting windows, improving engagement rates by 18% compared to baseline schedules. Developed an NLP-powered recommendation engine using transformer embeddings to cluster high-performing posts and suggest tailored content ideas, reducing creation time for marketing teams by 25%. Implemented an analytics dashboard and RESTful API (FastAPI + Plotly) delivering real-time campaign performance insights and AI-driven suggestions, enabling faster iteration cycles for clients.
Limurse
Remote
Jan 2025 — Mar 2025
Built a predictive content engagement model with scikit-learn and XGBoost to forecast optimal posting windows, improving engagement rates by 18% compared to baseline schedules. Developed an NLP-powered recommendation engine using transformer embeddings to cluster high-performing posts and suggest tailored content ideas, reducing creation time for marketing teams by 25%. Implemented an analytics dashboard and RESTful API (FastAPI + Plotly) delivering real-time campaign performance insights and AI-driven suggestions, enabling faster iteration cycles for clients.
Swati Web Technologies
Bhopal, MP
Jun 2024 — Aug 2024
Engineered data pipelines using Pandas and NumPy to clean and preprocess historical inventory and sales time-series data. Developed an LSTM-based demand forecasting model in TensorFlow/Keras, reducing forecast error by 15% vs. an ARIMA baseline; used time-series cross-validation and Bayesian hyperparameter tuning to optimize predictive accuracy. Deployed the model as a RESTful API (FastAPI on Docker) and integrated predictions into an internal dashboard (Plotly Dash); improved forecast accuracy by 20%, significantly reducing stockouts and enabling data-driven replenishment decisions.
Swati Web Technologies
Bhopal, MP
Jun 2024 — Aug 2024
Engineered data pipelines using Pandas and NumPy to clean and preprocess historical inventory and sales time-series data. Developed an LSTM-based demand forecasting model in TensorFlow/Keras, reducing forecast error by 15% vs. an ARIMA baseline; used time-series cross-validation and Bayesian hyperparameter tuning to optimize predictive accuracy. Deployed the model as a RESTful API (FastAPI on Docker) and integrated predictions into an internal dashboard (Plotly Dash); improved forecast accuracy by 20%, significantly reducing stockouts and enabling data-driven replenishment decisions.
Swati Web Technologies
Bhopal, MP
Jun 2024 — Aug 2024
Engineered data pipelines using Pandas and NumPy to clean and preprocess historical inventory and sales time-series data. Developed an LSTM-based demand forecasting model in TensorFlow/Keras, reducing forecast error by 15% vs. an ARIMA baseline; used time-series cross-validation and Bayesian hyperparameter tuning to optimize predictive accuracy. Deployed the model as a RESTful API (FastAPI on Docker) and integrated predictions into an internal dashboard (Plotly Dash); improved forecast accuracy by 20%, significantly reducing stockouts and enabling data-driven replenishment decisions.
Bachelor of Technology in Electronics & Computer Engineering
Mahindra University
2022 - 2026
Coursework includes Introduction to Computing, Machine Learning with Python, Data Structures & Algorithms, Object Oriented Programming, Operating Systems, Database Management Systems, and Computer Networks.
Bachelor of Technology in Electronics & Computer Engineering
Mahindra University
2022 - 2026
Coursework includes Introduction to Computing, Machine Learning with Python, Data Structures & Algorithms, Object Oriented Programming, Operating Systems, Database Management Systems, and Computer Networks.
Bachelor of Technology in Electronics & Computer Engineering
Mahindra University
2022 - 2026
Coursework includes Introduction to Computing, Machine Learning with Python, Data Structures & Algorithms, Object Oriented Programming, Operating Systems, Database Management Systems, and Computer Networks.
Senior Secondary Education - Science Stream
Allen Career Institute
2019 - 2021
I focused on Physics, Chemistry, and Mathematics, where I developed strong problem-solving skills, sharpened my analytical thinking, and built a solid foundation by preparing for competitive exams and tackling challenging quantitative problems.
Senior Secondary Education - Science Stream
Allen Career Institute
2019 - 2021
I focused on Physics, Chemistry, and Mathematics, where I developed strong problem-solving skills, sharpened my analytical thinking, and built a solid foundation by preparing for competitive exams and tackling challenging quantitative problems.
Senior Secondary Education - Science Stream
Allen Career Institute
2019 - 2021
I focused on Physics, Chemistry, and Mathematics, where I developed strong problem-solving skills, sharpened my analytical thinking, and built a solid foundation by preparing for competitive exams and tackling challenging quantitative problems.
Real-Time Social Context Audio Classifier
Self Initiated Project
Jun 2025 — Jul 2025
Built and deployed a real-time Streamlit app that classifies 3s audio clips into 5 social contexts using a compact CNN with log-Mel spectrograms; trained with early stopping and exported a production-ready checkpoint. Engineered a scalable preprocessing pipeline on Kaggle/GCS: parsed AMI annotations, generated 3s sliding-window clips (IHM/SDM), and synthesized “Noisy Environment” via MiniLibriMix augmentation, producing 17,997 labeled samples. Achieved 76% test accuracy across five classes; delivered confusion matrix and per-class metrics, and hardened the app with robust model checks, consistent feature shapes, and controlled inference latency
Real-Time Social Context Audio Classifier
Self Initiated Project
Jun 2025 — Jul 2025
Built and deployed a real-time Streamlit app that classifies 3s audio clips into 5 social contexts using a compact CNN with log-Mel spectrograms; trained with early stopping and exported a production-ready checkpoint. Engineered a scalable preprocessing pipeline on Kaggle/GCS: parsed AMI annotations, generated 3s sliding-window clips (IHM/SDM), and synthesized “Noisy Environment” via MiniLibriMix augmentation, producing 17,997 labeled samples. Achieved 76% test accuracy across five classes; delivered confusion matrix and per-class metrics, and hardened the app with robust model checks, consistent feature shapes, and controlled inference latency
Real-Time Social Context Audio Classifier
Self Initiated Project
Jun 2025 — Jul 2025
Built and deployed a real-time Streamlit app that classifies 3s audio clips into 5 social contexts using a compact CNN with log-Mel spectrograms; trained with early stopping and exported a production-ready checkpoint. Engineered a scalable preprocessing pipeline on Kaggle/GCS: parsed AMI annotations, generated 3s sliding-window clips (IHM/SDM), and synthesized “Noisy Environment” via MiniLibriMix augmentation, producing 17,997 labeled samples. Achieved 76% test accuracy across five classes; delivered confusion matrix and per-class metrics, and hardened the app with robust model checks, consistent feature shapes, and controlled inference latency
Contract Clause Analyzer - Intelligent Legal Comparison Agent
Self Initiated Project
Sept 2025 — Oct 2025
Developed a hybrid rules-LLM clause detection system that automatically classified 144 contract spans into 8 procurement-critical categories (Data Sharing, Termination for Convenience, Audit Rights, etc.), achieving consistent taxonomy mapping aligned to CUAD legal standards while extracting normalized attributes (notice days, termination fees, auto-renewal flags) for structured vendor comparison. Implemented a pgvector-backed semantic retrieval pipeline in PostgreSQL that ingests PDF contracts via layout-aware parsing (Unstructured), stores 384-dimensional clause embeddings with metadata filtering, and enables sub-second hybrid search across vendor agreements with span-level citations for legal transparency and compliance auditing. Engineered a transparent favorability scoring rubric that ranks vendor contracts by termination flexibility, liability caps, and audit rights using weighted attribute formulas, validated through LlamaIndex faithfulness/relevancy evaluators to ensure all answers remain grounded in original contract text spans.
Contract Clause Analyzer - Intelligent Legal Comparison Agent
Self Initiated Project
Sept 2025 — Oct 2025
Developed a hybrid rules-LLM clause detection system that automatically classified 144 contract spans into 8 procurement-critical categories (Data Sharing, Termination for Convenience, Audit Rights, etc.), achieving consistent taxonomy mapping aligned to CUAD legal standards while extracting normalized attributes (notice days, termination fees, auto-renewal flags) for structured vendor comparison. Implemented a pgvector-backed semantic retrieval pipeline in PostgreSQL that ingests PDF contracts via layout-aware parsing (Unstructured), stores 384-dimensional clause embeddings with metadata filtering, and enables sub-second hybrid search across vendor agreements with span-level citations for legal transparency and compliance auditing. Engineered a transparent favorability scoring rubric that ranks vendor contracts by termination flexibility, liability caps, and audit rights using weighted attribute formulas, validated through LlamaIndex faithfulness/relevancy evaluators to ensure all answers remain grounded in original contract text spans.
Contract Clause Analyzer - Intelligent Legal Comparison Agent
Self Initiated Project
Sept 2025 — Oct 2025
Developed a hybrid rules-LLM clause detection system that automatically classified 144 contract spans into 8 procurement-critical categories (Data Sharing, Termination for Convenience, Audit Rights, etc.), achieving consistent taxonomy mapping aligned to CUAD legal standards while extracting normalized attributes (notice days, termination fees, auto-renewal flags) for structured vendor comparison. Implemented a pgvector-backed semantic retrieval pipeline in PostgreSQL that ingests PDF contracts via layout-aware parsing (Unstructured), stores 384-dimensional clause embeddings with metadata filtering, and enables sub-second hybrid search across vendor agreements with span-level citations for legal transparency and compliance auditing. Engineered a transparent favorability scoring rubric that ranks vendor contracts by termination flexibility, liability caps, and audit rights using weighted attribute formulas, validated through LlamaIndex faithfulness/relevancy evaluators to ensure all answers remain grounded in original contract text spans.
AI-Powered Health Assistance
Self Initiated Project
Mar 2025 — Apr 2025
Developed a full-stack Flask app analyzing user symptoms against a 40,000+ medicine database, using a Hugging Face NLP model and a custom 50-condition knowledge base. Engineered a recommendation engine using TF-IDF and Cosine Similarity, identifying therapeutically equivalent medicine alternatives with over 95% accuracy based on composition. Implemented a prescription analysis feature using the Google Gemini Vision API to automate medicine name extraction from user-uploaded images.
AI-Powered Health Assistance
Self Initiated Project
Mar 2025 — Apr 2025
Developed a full-stack Flask app analyzing user symptoms against a 40,000+ medicine database, using a Hugging Face NLP model and a custom 50-condition knowledge base. Engineered a recommendation engine using TF-IDF and Cosine Similarity, identifying therapeutically equivalent medicine alternatives with over 95% accuracy based on composition. Implemented a prescription analysis feature using the Google Gemini Vision API to automate medicine name extraction from user-uploaded images.
AI-Powered Health Assistance
Self Initiated Project
Mar 2025 — Apr 2025
Developed a full-stack Flask app analyzing user symptoms against a 40,000+ medicine database, using a Hugging Face NLP model and a custom 50-condition knowledge base. Engineered a recommendation engine using TF-IDF and Cosine Similarity, identifying therapeutically equivalent medicine alternatives with over 95% accuracy based on composition. Implemented a prescription analysis feature using the Google Gemini Vision API to automate medicine name extraction from user-uploaded images.
Looking to start a project or you need consultation? Feel free to contact me.
Hyderabad, TN, India