Portfolio - Syed Rabbani

Syed Rabbani

AI Engineer

Resume

2

Machine Learning Projects

1

Deep Learning Project

2

GenAI Projects

About Me

Software engineer with over 2 years of experience in building and deploying AI solutions using machine learning and Generative AI. Passionate about extracting insights from large data, and deploying ML models that solve real-world problems.

Skills:

Programming Languages: Python, C/C++, SQL
ML & AI: Machine Learning, Deep Learning, NLP, LLMs, Semantic Search, Vector Databases, Knowledge Graphs
Libraries & Frameworks: Pandas, NumPy, Scikit-learn, PyTorch, Matplotlib, Seaborn, Spacy, HuggingFace, LangChain, Chromadb
Academics: Object Oriented Design, Data Structures & Algorithms, Agile Methodologies
Tools: PyCharm, VS Studio, Git, Streamlit, Jupyter Notebook, MLFlow

Area of Expertise:

➡️Building models that predict future trends and improve business decisions.
➡️Business understanding & Critical thinking to frame and solve data-driven problems.
➡️Strong foundation in mathematics and statistics to understand model behavior, optimize performance, and interpret results effectively.

Key Skills

Python

SQL

Pandas

Matplotlib

Numpy

Machine Learning

NLP

Deep Learning

GenAI

My Projects

Health Insurance Premium Prediction

Domain/Function: Insurance

Credit Risk Modelling

Domain/Function: Finance

Car Damage Detection using Deep Learning

Domain/Function: Image Classification

News research Assistant using RAG

Domain/Function: Text Generation

E-Commerce AI Chatbot

Domain/Function: E-Commerce

My Experience

Software engineer - Intel, Bangalore

July 2022 - May 2024

Developed ML-based memory prediction engine for EDA(Electronic Design Automation) jobs, improving resource utilization and reducing compute wastage by 45%.
Built an NLP-based ticket suggestion system using semantic search over historical data, improving issue resolution speed by 30% through dynamic identification of similar technical issues
Developed an AI chat application with LLM integration, achieving 85% accuracy in resolving customer queries.
Reduced resolution time by 40%, enabling faster, more accurate, and contextually relevant AI-generated responses.

Skills: Python

Awards & Certificate

Python: Beginner to Advanced For Data Professionals

SQL for Data Science

Math and Statistics For AI, Data Science

Master Machine Learning for Data Science & AI: Beginner to Advanced

Deep Learning: Beginner to Advanced

Natural Language Processing

Gen AI to Agentic AI with Business Projects

Let's Connect

Feel free to get in touch with me. I am always open to discussing new projects, creative ideas or opportunities to be part of your visions.

Download Resume

Resume

Mail Me

[email protected]

https://www.linkedin.com/in/rabbani-syed-13a7801b1/

Call Me

+91 9603206894

GitHub

https://github.com/rabbani104

Machine Learning Project

Health Insurance Premium Prediction

Domain/Function: Insurance

SHIELD Insurance faced declining market share due to Inadequate insurance pricing. Tasked with predicting the insurance cost for individuals, Used Machine Learning models for predicting insurance premium with 98% accuracy. ML models helped to gain 25% market share by accurate cost predictions.

View Project
Engagement

View Project
on GitHUB

View Project
on

Feature importance analysis revealed that insurance plan, genetic risk, and risk score have the highest impact on predicting insurance costs.

At least 95% of prediction errors have a percentage difference of less than 10% between the predicted and actual values.

Project Details

SHIELD Insurance faced a declining market share due to inaccurate premium pricing, causing customer loss and financial inefficiencies. The company needed a data-driven solution to optimize pricing and stay competitive.

The goal was to develop a Machine Learning model to predict insurance costs accurately based on customer demographics and health factors, ensuring competitive and risk-adjusted pricing.

Collected and cleaned historical insurance data, performed EDA to identify key factors, and built Linear Regression and XGBoost models. Tuned hyperparameters and evaluated models using R² Score achieving 98% accuracy. Deployed model using streamlit for real-time premium predictions.

The ML model improved pricing accuracy, leading to a 25% increase in market share. Automated premium predictions enhanced efficiency, reducing manual errors and boosting customer retention.

Machine Learning Project

Credit Risk Modelling

Domain/Function: Finance

Lauki Finance, an NBFC, faced rising loan defaults and approached me to build a credit risk model to predict default probability, credit score, and rating. Built various ML models(Logistic Regression, Random Forest, and XGBoost), LR achieved 95% accuracy, 96% recall, and reduced defaults by 15%.

View Project
on GitHUB

View Project
on

Graphs showing feature distribution of default and non-default.

Picture depicting AOC curve

Project Details

Lauki Finance, an NBFC, faced rising loan defaults, causing financial losses. They needed a credit risk model to assess default probability, credit scores, and ratings using factors like age, income, loan amount, and loan-to-income ratio.

The goal was to develop a Machine Learning model that could accurately predict default risk, helping the company make informed lending decisions and reduce defaults.

Collected and cleaned historical loan data, performed EDA to identify key risk factors, and built Logistic Regression, Random Forest, and XGBoost models. Tuned hyperparameters, achieving 95% accuracy and 96% recall. Integrated the model into Lauki Finance’s system via streamlit for real-time risk assessment.

The ML model led to a 15% decrease in loan defaults, improved credit risk evaluation, and enhanced lending efficiency, boosting financial stability.

Deep Learning Project

Car Damage Detection using Deep Learning

Domain/Function: Image Classification

Developed an AI model for Vrooms to classify car damage into 6 categories (e.g., Front Crushed). Used CNN, then improved accuracy with transfer learning (EfficientNet, ResNet50) and Optuna tuning (80% accuracy). Deployed via Streamlit, enabling fast damage assessment for pricing and insurance claims

View Project
Engagement

View Project
on GitHUB

View Project
on

User uploads a Front Breakage image, and the model correctly classifies it as Front Breakage.

User uploads a Rear Crushed image, and the AI model processes it, analyzes features, and accurately classifies it as Rear Crushed.

Project Details

Vrooms, a car company, sought an automated system to detect and classify exterior car damage accurately. Manual inspections were time-consuming and prone to inconsistencies, impacting used car pricing and insurance claim processing.

Developed a model to classify car images into six categories (Front/Rear Normal, Crushed, Breakage).

Built a CNN model, improved accuracy using regularization. To improve accuracy applied transfer learning with EfficientNet & ResNet50, then optimized hyperparameters using Optuna, achieving 80% accuracy. Deployed the model as a Streamlit web app, allowing users to upload images and get instant damage classification.

Automated damage assessment, eliminating manual inspections by 75%. This accelerated used car pricing and improved efficiency for P&C insurers by enabling faster claim settlements by 45%, enhancing customer satisfaction.

GenAI Project

News research Assistant using RAG

Domain/Function: Text Generation

This project is designed to help users extract relevant information from news articles based on provided URLs. The user inputs three URLs and asks a question related to the content within those links. The system employs a Retrieval-Augmented Generation (RAG) approach to fetch and synthesize answers.

View Project
on GitHUB

View Project
on

Other Tools

Python

LangChain

HuggingFace

ChromaDB

User Interface

UI showing Question/answer content

Project Details

With the vast amount of news available online, users often struggle to extract relevant information from multiple sources efficiently. Manually reading and summarizing articles is time-consuming and ineffective for quick research.

Implemented LangChain for structuring the retrieval and response generation process. Utilized Hugging Face models to process and generate human-like answers. Employed ChromaDB to store and retrieve vector embeddings for efficient information retrieval. Designed a system where the user inputs atleast one URL and asks a question related to the articles. The model retrieves relevant information from the stored vector database and generates a precise response.

Successfully built an AI-powered research assistant that automates news content analysis. Enabled users to get direct, concise answers without manually reading entire articles. Improved efficiency in extracting relevant information with 90% accuracy, making news research faster and more effective.

GenAI Project

E-Commerce AI Chatbot

Domain/Function: E-Commerce

This is POC of an intelligent chatbot tailored for an e-commerce platform, enabling seamless user interactions by accurately identifying the intent behind user queries. It leverages real-time access to the platform's database, allowing it to provide precise and up-to-date responses.

View Project
on GitHUB

View Project
on

Other Tools

Python

Chroma

Streamlit

Semantic-Router

Groq

Chromadb

Pandas

sqlite

Chatbot answering FAQs

Chatbot retrieving products based on user requirement

Project Details

E-commerce platforms often face challenges in handling diverse customer queries efficiently. Manual responses lead to delays and inaccuracies, impacting user experience. To address this, an AI-powered chatbot was needed to provide real-time, accurate, and context-aware responses.
Developed an intelligent chatbot capable of understanding user intent and providing real-time responses by leveraging Groq, LLaMA LLM, ChromaDB, and SQLite.

Implemented Groq-powered LLaMA LLM for intent classification and natural language understanding, enabling the chatbot to handle:
-> FAQ queries – Answering platform-related questions (e.g., payment options).
-> SQL-based queries – Fetching real-time product listings using SQLite.
-> Small Talk – Engaging in casual conversations for enhanced user experience.

Automated instant and accurate responses, reducing human intervention by 35%. Boosted customer satisfaction and streamlined support automation by 50% for the e-commerce platform.