► Architected and built a novel "source-first" Retrieval-Augmented Generation (RAG) system for a generative AI research platform, completely eliminating model hallucinations and guaranteeing 100% factual integrity by grounding LLM outputs in pre-verified sources.

► Slashed a critical data collection workflow from weeks to just hours by engineering a multi-stage automation tool using spaCy and LLMs, resolving a major project bottleneck and accelerating research output for a published academic paper.

► Developed and deployed interactive Plotly and Dash dashboards to visualize emerging semantic trends from 550+ academic publications, enabling researchers to dynamically explore complex topic models and word co-occurrence patterns.

► Operationalized a high-frequency stock prediction model by architecting and deploying a complete, end-to-end MLOps pipeline on AWS SageMaker. The system featured a full CI/CD workflow for automated model retraining, versioning, and deployment to a low-latency API endpoint for real-time inference.

► Automated over 90% of manual data entry for a key enterprise client by building and deploying a multi-modal AI service on Azure that fused OCR and BERT-based NLP to intelligently extract and structure data from complex tender documents.

► Partnered directly with cross-functional client teams to translate ambiguous business needs into concrete technical requirements, manage the project lifecycle, and deliver production-ready NLP models.

► Built and deployed a high-throughput, containerized RESTful API using FastAPI and Docker to serve a resume parsing model, architecting the system to handle thousands of concurrent requests for an enterprise-scale recruitment platform.

► Boosted entity extraction accuracy from 78% to 91% F1-score by fine-tuning and integrating a state-of-the-art Hugging Face Transformer model for Named Entity Recognition (NER), significantly improving the quality of automated candidate screening.