Senior AI Engineer

Architected and implemented a multi-agent orchestration framework using LangGraph, enabling specialised agents for market analysis, risk assessment, portfolio evaluation, and trade recommendation. Developed memory-enabled agent workflows that incorporated historical trade outcomes, portfolio exposure, and previous reasoning traces to improve decision consistency across trading sessions. Built Event-based Retrieval-Augmented Generation (RAG) pipelines using OpenSearch and hybrid retrieval to provide agents with historical price patterns, earnings reports, macroeconomic events, research documents, and trading playbooks during decision making. Implemented reflection and validation workflows that evaluated generated trade recommendations against predefined risk rules and historical outcomes, enabling iterative refinement before execution. Developed asynchronous agent communication and orchestration pipelines using FastAPI and asyncio, reducing end-to-end analysis latency for concurrent stock evaluations. Engineered an end-to-end deep learning pipeline to train a Node Transformer, utilizing attention mechanisms to capture complex dependencies in large-scale relational datasets. Integrated LLM reasoning with quantitative models, technical indicators, sentiment analysis, and portfolio risk metrics to generate explainable trade recommendations instead of standalone model predictions. Added production-grade visibility using LangSmith and structured logging to trace multi-agent execution, debug reasoning paths, and evaluate agent performance for every execution. Collaborated with quantitative researchers and product stakeholders to integrate AI decision-support workflows into the existing trading platform while maintaining a modular agent architecture.

Created GenAI-powered claim summarization services that converted lengthy medical records and claim histories into concise risk assessment reports, reducing analyst review time by 60%. Built Retrieval-Augmented Generation (RAG) pipelines to retrieve relevant historical claims, policy guidelines, medical procedures, and regulatory documents for risk assessment workflows. Developed natural language search capabilities allowing insurance analysts to query claims data using business questions instead of complex SQL queries.

Designed and deployed a GenAI-powered validation assistant leveraging Retrieval-Augmented Generation (RAG) to analyze firmware requirements, test specifications, execution logs, and historical defects. Enabled automated test-case generation, requirement-to-test traceability, coverage analysis, and intelligent failure investigation. Built embedding-generation and indexing pipelines using transformer-based models, reducing information retrieval latency from several minutes of manual search to under 5 seconds. Engineered low-latency inference APIs capable of serving validation insights with average response times below 2 seconds while supporting concurrent engineering teams. Contributed to the deployment of production-grade AI services with monitoring, logging, and feedback loops, achieving 99.5% service availability for validation teams.

Architected and implemented model training pipelines and inference frameworks for ANPR and 3D Traffic Light applications in Intelligent Traffic Management Systems. Optimized model lifecycle workflows including dataset preparation, training, evaluation, model versioning, and deployment. Delivered production-ready AI services that enabled seamless integration of computer vision capabilities with backend traffic management platforms. Reduced Average Wait Time at traffic signals by 15% through intelligent mobility solutions, resulting in daily cost savings of approximately $300.

Hands-on expertise in training, fine-tuning and deploying Object detection models using deep learning in AWS and GCP environments related to advanced driver assistance systems, including image and predictive modelling. Tools: Python, NumPy, Pandas, PyTorch, TensorFlow, LabelImg, AWS (EC2, S3, SageMaker), Docker, SQL, PostgreSQL