Title of the talk
Practical Machine Learning in Real-World Applications
Description
Hi,
I would like to propose a talk for upcoming event.
🧠 Topic:
Practical Machine Learning in Real-World Applications
🎯 What I’ll Cover:
How machine learning works beyond theory
Real-world project examples (classification, prediction)
Common challenges in ML projects (data quality, overfitting, deployment issues)
Tools & technologies used (Python, scikit-learn, etc.)
Tips for beginners to get started in ML
💡 Why This Talk:
Many people learn ML theoretically but struggle to apply it practically. This talk will bridge that gap by sharing real-world insights and practical approaches.
⏱️ Duration:
~20–30 minutes
Looking forward to contributing and sharing my experience!
Thanks 🙌
Table of contents
Built an end-to-end real-time fraud detection pipeline for a fintech client processing 2.3 million transactions daily. The existing rule-based system was generating high false
positives and missing sophisticated fraud, costing ₹4.2 crore per quarter.
Replaced it with an XGBoost model trained on 100K transactions, deployed behind a FastAPI endpoint capable of returning a fraud probability score and an approve / review / block
decision — all under 200ms. Transactions stream through Apache Kafka, are processed in micro-batches via PySpark Structured Streaming, and results are visualized on a live
Streamlit dashboard. Experiments tracked end-to-end with MLflow. The entire infrastructure is containerized with Docker Compose.
Targets met: Precision > 90%, Recall > 85%, False Positive Rate < 5%.
This covers the problem → solution → stack → outcome arc — which maps cleanly to a talk outline like:
- The Problem (rule-based system failures)
- The Approach (ML model + streaming pipeline)
- The Architecture (Kafka → Spark → XGBoost → API)
- Results & Metrics
Duration (including Q&A)
20-30 minutes
Prerequisites
No response
Speaker bio
Rushikesh Panjabrao Chavan is an AI/ML Engineer at NexGenesis with a strong foundation in Machine Learning, Deep Learning, and Generative AI. He holds a B.Sc. in Physics from
Bharati Vidyapeeth University, Pune, and has self-driven his way into building production-grade AI systems.
Core expertise:
- GenAI & LLMs — RAG systems, LangChain, Hugging Face, Prompt Engineering, Gemini
- ML/DL — Scikit-learn, TensorFlow, Keras, PyTorch
- MLOps — MLflow, Prefect, Flask, Streamlit, GCP, ChromaDB, PySpark
Hands-on projects include:
- A RAG system for research document Q&A (FinVeda RAG project visible in the directory)
- Fraud detection, Predictive Maintenance, PharmaCast — real-world ML pipelines
- CNN-based Rice Grain Classification (96% accuracy), Telecom Churn Prediction, Flipkart Sentiment Analysis
Beyond code, he's an open-source contributor to scikit-learn and Hugging Face, and has published technical articles on Medium covering language modeling and transformers.
The talk/workshop speaker agrees to
Title of the talk
Practical Machine Learning in Real-World Applications
Description
Hi,
I would like to propose a talk for upcoming event.
🧠 Topic:
Practical Machine Learning in Real-World Applications
🎯 What I’ll Cover:
How machine learning works beyond theory
Real-world project examples (classification, prediction)
Common challenges in ML projects (data quality, overfitting, deployment issues)
Tools & technologies used (Python, scikit-learn, etc.)
Tips for beginners to get started in ML
💡 Why This Talk:
Many people learn ML theoretically but struggle to apply it practically. This talk will bridge that gap by sharing real-world insights and practical approaches.
⏱️ Duration:
~20–30 minutes
Looking forward to contributing and sharing my experience!
Thanks 🙌
Table of contents
Built an end-to-end real-time fraud detection pipeline for a fintech client processing 2.3 million transactions daily. The existing rule-based system was generating high false
positives and missing sophisticated fraud, costing ₹4.2 crore per quarter.
Replaced it with an XGBoost model trained on 100K transactions, deployed behind a FastAPI endpoint capable of returning a fraud probability score and an approve / review / block
decision — all under 200ms. Transactions stream through Apache Kafka, are processed in micro-batches via PySpark Structured Streaming, and results are visualized on a live
Streamlit dashboard. Experiments tracked end-to-end with MLflow. The entire infrastructure is containerized with Docker Compose.
Targets met: Precision > 90%, Recall > 85%, False Positive Rate < 5%.
This covers the problem → solution → stack → outcome arc — which maps cleanly to a talk outline like:
Duration (including Q&A)
20-30 minutes
Prerequisites
No response
Speaker bio
Rushikesh Panjabrao Chavan is an AI/ML Engineer at NexGenesis with a strong foundation in Machine Learning, Deep Learning, and Generative AI. He holds a B.Sc. in Physics from
Bharati Vidyapeeth University, Pune, and has self-driven his way into building production-grade AI systems.
Core expertise:
Hands-on projects include:
Beyond code, he's an open-source contributor to scikit-learn and Hugging Face, and has published technical articles on Medium covering language modeling and transformers.
The talk/workshop speaker agrees to
Share the slides, code snippets and other material used during the talk
If the talk is recorded, you grant the permission to release
the video on PythonPune's YouTube
channel
under CC-BY-4.0
license
Not do any hiring pitches during the talk and follow the Code
of
Conduct