Sadiq Pasha

Client - Deutsche Bank.

• Developed conversational multi-agents with RAG pipelines using Autogen Framework for Vendor Risk Management use case, thereby reducing the time of onboarding Vendors from 15 days to 1 day, saving the client more than $1.5M per annum.
• Designed and led the development of a data exfiltration system using a hybrid approach (ML and rule-based), utilizing log-based metadata generated by multiple GCP services, with scalable Vertex AI Pipelines, cloud logging, and Big Query as a data sink service. This solution could potentially save millions of dollars for bank in fines for data breach incidents.
• Documented technical specifications, functional specs, and architecture design records for multiple GCP services thoroughly.
• Supervised and evaluated team performance each day, and facilitated educational sessions for less experienced colleagues.
• Facilitated recruitment process to onboard competent individuals in data science and ML engineering roles.

Clayfin's PFM (Personal Finance Management) App

• Developed a scalable script to extract information like Merchant details, Bank Charges and transaction details from Finvu's extensive multi-bank datasets that was needed to auto-categorize the transactions.
• Developed a multi-step ML model for auto-categorization of transactions. Steps mainly included using MCC(Merchant Category Code), Keywords from bank statements, Text Entity Extraction, Customer Segmentation using Clustering algorithms and an user preference option to categorize the transactions.
• Developed customer churn prediction, customer retention with Cohorts Analysis for time, size and location of users & customer scoring models using RFMT Analysis and xgboost model
• Developed Models and APIs for Expense forecasting, Monthly Balance Prediction that were integrated into PFM mobile app.
• Fincast model for budget prediction using ARIMAX/fbprophet models to quarterly predict bank balance
• Developed customized insights like top merchant spendings, change in spendings, Salary change, similar subscriptions, unusual income deposits, investment opportunities etc, that were displayed on a monthly basis on the app.

Client - Ab inBev

• Developed multiple hybrid(ML+Rulebased) explainable, scalable fraud detection models for procure-to-pay(P2P) usecase.
• Reducing both False Positives and False Negatives and enhanced F1 score from 59% to 86% with Cross-validation techniques, active learning for auto-labelling and using an advanced model like Auto-encoder model

Client - Unilever Indonesia

• Developed outlet segmentation model to detect underperforming outlets and devised a strategy to improve the sales ratio for the outlets. Strategy included detecting peak hours and offering loyalty benefits. Reduced losses by 17% for the quarter.

Personal Chase Credit Card

• Built Customer Segmentation model to effectively market Chase credit cards to diverse customers based on their spending and income habits.

Mytechub chatbot

• Developed a Dialogflow CX chatbot as a service offered to internal chase customers to offer assistance on various issues and knowledge bases. Also deployed on mytechub mobile application, reduced GTI incidents by 85%, thus saving bank millions of dollars in savings