Cherupally S

AI / ML Engineer with 3+ years building AI-powered applications, machine learning models and generative AI features that run in production.

At First National

Bank I train the models, build the data pipelines behind them, wrap them in REST APIs and deploy them as containers through CI/CD on GCP Vertex AI and Azure ML. My stack is Python with PyTorch and TensorFlow for modelling, LangChain and vector search for the LLM work, and Docker plus GitHub Actions for delivery. I live in Midrand.

AI Engineer - First National Bank (FNB) South Africa, FirstRand Group - Fairland Campus, Johannesburg, South Africa

(2023-07)

• Built SIM-swap fraud detection using XGBoost and Isolation Forest on engineered telecom / transaction features (OTP spikes, device changes, impossible travel, login shifts) to score high-risk accounts before fund loss.
• Designed an LLM GenAI reasoning layer with a RAG pipeline over financial and regulatory documents using LangChain, vector embeddings and ChromaDB, integrating model scores, retrieved document evidence and identity signals into structured case summaries with deterministic validation checks to reduce hallucination risk.
• Built a document understanding pipeline using OCR, layout parsing and Vision-Language Models (VLMs) for automated classification, field extraction and key-value parsing across identity and financial documents.
• Engineered time-aware, real-time feature pipelines from telecom, authentication and transactional logs with drift detection, monitoring and analyst feedback loops for continuous fraud scoring.
• Orchestrated fraud and identity workflows on GCP Vertex AI Pipelines and Azure (Azure ML, Logic Apps, Azure AI Foundry), automating training, batch scoring and monitoring while enforcing security controls and regulatory auditability.
• Developed a step-up identity verification pipeline using computer vision and multimodal AI for document classification, tamper detection and liveness-enabled face matching on high-risk accounts.

Machine Learning Intern, Intel Unnati Industry Training Programme - Intel Corporation - Remote, India

(2023-05 - 2023-07)

• Built a binary fake news classification system, executing the full ML lifecycle from EDA and preprocessing to model training, evaluation and deployment.
• Engineered an end-to-end NLP preprocessing pipeline with TF-IDF, stop word removal, stemming and tokenization, converting raw news text into numerical features for supervised classification.
• Designed and deployed an interactive Streamlit web application as the inference interface, delivering real-time predictions and class probability scores via a Python backend.

B.Tech - Artificial Intelligence & Machine Learning - Guru Nanak Institute of Technology, JNTU (2019-12 - 2023-05)