Talha Zuberi

Manager AI

Askari Bank • Full-time

Nov 2025 - Present • 6 mos

• Leading AI initiatives for enterprise banking automation, NLP-based chatbots, and compliance tools. • Voice Bot — AI voice banking MVP enabling fund transfers, bill payments, balance checks via natural voice commands. • IS Bot — InfoSec compliance RAG system analyzing documents across multiple frameworks. • Islamic Bot — Multi-agent RAG chatbot for Shariah-compliant banking products. • Askari Buddy — Multi-modal banking chatbot (text + voice), multi-lingual, 9 LangChain agents, under active development. • Compliance and TBML Automation — Financial news scraping, TBML variance reporting, and PDF conversion for regulatory compliance.

AI Engineer

Horizon Tech Services • Full-time

Mar 2024 - Nov 2025 • 1 yr 8 mos

Designed and deployed scalable anomaly detection models for User Entity and Behavior Analytics (UEBA) with Security Information and Event Management (SIEM) system, improving threat detection accuracy by 35% and reducing false positives by 40%. • Developed a real-time log correlation system processing 500,000 events per second from NDR and EDR streams, increasing sophisticated attack detection by 28% and streamlining data ingestion and analysis. • Leveraged correlated logs from multiple UEBA use cases and implemented multiclass classification techniques to detect lateral movement patterns across enterprise endpoints • Achieved a 95% accuracy rate in classifying malicious PDF, PE, and Office files using machine learning models, enhancing overall threat detection capabilities by 50% through static and dynamic techniques. • Developed a multiclass malware classification model for PDF, PE, and Office files using antivirus labels from multiple defender software. The model was deployed, including the ML classifier and 1D CNN, within a custom-built antivirus solution. • Led the development of NLP and DGA detection models using deep learning, contributing to behavior-based threat modeling. • Converted various types of files using a custom dumper to generate memory dumps. Applied NLP techniques, ML and DL models to analyze and classify memory dumps as malicious or benign. • Developed a standalone .exe for the deployment of machine learning models, allowing seamless execution on Windows systems to ensure cross-environment compatibility, simplified distribution and reduced configuration complexity for end users. • Integrated AI pipelines on Ubuntu servers using Docker, Kafka, Opensearch, and Apache NiFi.

Research Associate

Smart AgriTech • Full-time

Jan 2023 - Feb 2024 • 1 yr 1 mo

Build a Facial Recognition System for RIC National University of Science and Technology (NUST)