Yeshvanth Prabakar

Site Reliability Engineer @BNY

London, GB

Incident Dashboard - Architected and delivered a production-grade GenAI web application for incident classification and trend analysis using transformer-based LLMs (LangChain, custom NLP pipelines, Python SDKs), significantly improving operational insight and automation. This delivered a ~10% reduction in incident volume across key asset servicing applications within a few months of pilot- Built and maintained scalable data pipelines for ingestion, transformation, and storage of unstructured operational datasets, improving data readiness and enabling real-time analysis across distributed systems- Partnered closely with model-risk and compliance teams to standardise secure data handling, building controls for LLM privacy, redaction, and regulatory alignment—critical for AI usage in sensitive asset servicing domains- Tuned LLM workflows through prompt engineering, retrieval optimisation, and model drift controls, improving classification accuracy by 15% and strengthening the reliability of AI-driven insights.Vulnerability-EOL AI Correlation - Spearheaded a project under the industry leading AI-Hub practice to correlate end of life software to vulnerabilities- These datasets traditionally operate in silos, and patching vulnerabilities does not always equate to running approved or up-to-date software versions- Developed a custom matching algorithm that combines TF-IDF vectorisation, cosine similarity, and LLM-based predictive insights to map software entries to relevant vulnerabilities. The algorithm supports continuous learning whilst maintaining a high degree of accuracy for matches- Addressed high false positive rates by building a dynamic correlation knowledge base, incorporating domain-specific mappings and feedback from remediation teams to improve the accuracy and maintainability of the matching engine- Actively involved in hallucination mitigation and validation workflows, ensuring that LLM-generatedcorrelations meet accuracy thresholds.