Career
My Journey
A timeline of roles, research, and the problems I've worked on.
2025-08 – Present
Senior AI Engineer
Santander Bank, UK · London
Santander Bank is a global financial institution providing retail, commercial, and corporate banking services across Europe and the Americas. As a Senior AI Engineer, I architect and deliver enterprise grade GenAI systems that automate financial intelligence, enhance customer analytics, and modernise operational workflows across the bank’s digital ecosystem. My work spans multi agent RAG systems, document intelligence, LLM driven classification, and cloud‑native orchestration integrated with Salesforce and internal banking platforms.
- Designed and built the Precision Prospect GenAI system that generates structured summaries of company annual reports using AWS Textract OCR, Python based financial analysis, and multi step prompt chaining techniques for progressive extraction, validation, and summarisation.
- Developed a scalable document processing pipeline integrating FullCircl annual, financial, and shareholder datasets, performing analytical transformations before passing enriched data into LLM workflows.
- Implemented the full workflow using AWS Lambda, AWS Step Functions, and Amazon S3, enabling event‑driven report generation triggered directly from Salesforce, with outputs stored as HTML reports.
- Built a comprehensive evaluation framework using DeepEval, combining qualitative LLM based assessment with quantitative comparison against FullCircl ground‑truth datasets to ensure accuracy and reliability.
- Architected an Agentic AI RAG chatbot using page‑indexing retrieval techniques, semantic chunking, metadata tagging, and multi agent orchestration to deliver high precision responses across large internal document repositories.
- Implemented robust retrieval pipelines using LLM based reasoning, agent routing logic, and fallback strategies to minimise hallucinations and improve response consistency.
- Developed a GenAI driven call transcript classification system mapping transcripts to predefined L1 and L2 categories using hybrid rule based preprocessing and LLM semantic classification.
- Evaluated model performance using automated test cases, human in the loop feedback, and continuous monitoring of classification accuracy and drift.
- Applied advanced prompting techniques, guardrails, logging, and modular fail‑safe architectures to mitigate AI risks and ensure compliance with Santander’s Responsible AI standards.
- Collaborated closely with cloud engineers, Salesforce teams, data engineers, and business stakeholders to design secure, scalable, and production ready AI solutions aligned with enterprise governance and regulatory requirements.
2024-03 – 2025-07
Senior AI Engineer/Data Scientist
Crown Commercial Services · London, UK
Crown Commercial Services is a UK Government agency responsible for providing commercial and procurement services to the public sector. At CCS, I designed and built an intelligent Retrieval Augmented Generation (RAG) based search engine to replace the CCS Guided Match process which is a manual step-by-step keyword navigation tool that took each user 3-4 minutes per simple query. I also developed an advanced chatbot system that helped answer employee queries regarding company policies, HR and operations. I incorporated all this into a wider AI Agents system with 5 specialized agents. This was built on the Azure Cloud Platform, and I exposed the system as an API through an Azure web app that I built using Flask and integrated it into the CCS Knowledge Centre and other related business platforms.
- The multi-agent AI system was built using OpenAI GPT 40 and LangGraph framework for orchestration.
- The supervisor agent used a 1D CNN model for coordination and decision making.
- Developed a RAG based document processing pipeline with extraction of different structures of data from different file formats using OCR and similar mechanisms, automated indexing workflow with semantic chunking of embeddings, metadata tagging and information retrieval using Azure AI Search (Azure Cognitive Search)
- OpenCV image processing was used for layout parsing in the documents such as table structure detection and pre-OCR enhancements
- Implemented BERT Topic modelling combined with the LLM to categorize and label customer complaints and populated a real-time dashboard for the customer service team.
- Evaluated model performance using automated test cases and with human in the loop feedback techniques.
- Built light weight chatbots using Azure with Flask and scalable FastAPI endpoints integrated with Azure AI Search and OpenAI LLMs.
- Experimented with OpenAI Agents SDK for better orchestration and agents implementation
- Setup CI/CD pipelines with GitHub Actions and Azure DevOps.
- Monitored the AI system performance using Azure Monitor and Application Insights tracking the health of the system, response times and error / fallback rates.
- Mitigated AI Risks by implementing different strategies like appropriate prompting techniques, guardrails, logging and fail-safe modular architectures.
- Collaborated with business stakeholders, MLOps engineers and cloud engineers for designing the scalable and compliant AI solution.