I lead the fraud detection team at Razorpay Technologies Pvt. Ltd. Razorpay is a payment gateway solution provider. It handles more than 2 million transactions daily. My role is to build, deploy and maintain models that have high precision and recall for identifying fraudulent transactions. 

I've built an ensemble of xgboost and lightgbm model to attain better model performance.
The deployed approach for marking a transaction is a combination of manual rules and ML model results to attain better performance in a short duration.

Senior Data Scientist

Built a NLP framework using Allennlp, pytorch in python. It can be used for wide variety of NLP related tasks ie. multilabel multiclass classification (labelling the defects in manufacturing batch after reading the review from the QA engineer), reading comprehension (understanding the contracts and then answering the user's questions). This framework is implemented using Flask API with mail notifications. The parameters of the models can be manually iterated by the user.

Data Science Freelancer

I completed 3 major projects and assisted with 8 minor projects during my tenure with Sears Holdings Pvt. Ltd.
1. Product Matching : Created sequential model using 2 CNNs & 1 NLP for Product Matching across various E-commerce websites. Achieved accuracy 99.5% across Sears, K-Mart, Target, Amazon, and Nordstrom website and 95%+ across others
2. Product Recommendation : Implemented CNN for feature extractions to enhance the existing Collaborative Filtering based product recommendation. Improved the F1 score by 60%
3. Machine/ Deep Learning Templates : Developed 14 templates for machine learning algorithms. Various functions such as one-hot encoding, train-test split, outlier detection, normalization, performance recording against standard metrics are automated.

Machine Learning Engineer

Team Computers provide high impact analytics to MNCs across the globe. I worked in Automobile & Retail domain with Asia Pacific clientele.

Solutions Architect

Sookshm is a data analytics consulting organization that works with small & medium manufacturing companies to help them compete and grow profitably. In the year 2014, Sookshm helped over 20 Small Medium Enterprises gain access to strategies that increase efficiency and promote growth. Resulting from the organization’s work, Sookshm's clients have realized a total positive economic impact of more than $ 2 million through cost reduction.

Business Partner

A few machine learning projects:
- Recommendation System: Developed a product recommendation system using normalized item-to-item collaborative filtering  for a US-based online retailer
- Customer Attrition: Developed a random forest-based predictive model to predict the customers with a high propensity to attire within the next 1 month. Predict 72% of customers accurately
- Order Forecasting: Built an order forecasting model for one of the largest online retailers in India using linear regression taking different variables into consideration such as Marketing campaigns, discounts, and Promotion Campaigns etc.
- Purchase Prediction System: Developed a neural net for predicting the next purchase date and category for a grocery customer using 4 layered neural net. The model showcased 76% accuracy


6 years of experience in Python for machine learning, data analysis, database connections and plotting data:
1. Partnered with Google Cloud Solutions, India: Worked with multiple Large Clients in Automotive, E-Commerce and Travel Sector in India to deliver solutions based on Python hosted at Google Cloud for data processing and data visualization.
2. Analytical Solution for UK Based Finance Company: Worked with a 10 member team to deliver analytics solution for the finance firm. The main challenge was processing of high volume data at high velocity to generate live reporting solutions.
3. Multiple small solutions for SMEs based in USA, Cananda and UK: Worked to develop multiple minor analytical solutions with tasks ranging from database connection, extraction of structured and unstructured data, trasnformation of data through Python, loading of data to online hosted databases

I have created 10+ applications on R for Data Cleaning, Data Analysis, NLP and Machine Learning on R. I started with development in R in Feb 2016 and have increasingly worked with various tools since then.

7+ years experienced Data Scientist with 2+ years experience in Deep Learning

Rahul Agarwal

Implemented CNN for feature extractions to enhance the existing collaborative filtering based product recommendation. 

Improved the F1 score by 60%

Image features extraction for enchancing product recommendation

Created sequential model using 2 CNNs & 1 NLP for Product Matching across various E-commerce websites.
Achieved accuracy 99.5% across Sears, K-Mart, Target, Amazon, and Nordstrom website and 95%+ across others

Matching competition's online catalogue with client's catalogue

Allennlp is an open-source NLP research library, built on PyTorch by Allennai. AllenNLP makes it easy to design and evaluate new deep learning models for nearly any NLP problem, along with the infrastructure to easily run them in the cloud or on a laptop.

I was working on enhancing the capability of allennlp by developing a multilabel algorithm along with custom new metrics for multi-label model evaluation such as overall and average multilabel f1, multilabel AUC etc.. I used free datasets from kaggle jigsaw comment classification and IMDB movie reviews dataset.

I've also tested it on the client's propriety email dataset and compared the results against the legacy - rule-based, text blob, nltk and spacy results. The customized allennlp showcased much better results against the previous algorithms

Personal details

Rahul A.

Summary

Technical skills

Work Experience

Personal Projects