Betsy جورج

• Architected and deployed production RAG systems using LangChain and LCEL — designing full
pipelines for document ingestion, adaptive chunking, embedding generation, FAISS/Chroma
vector indexing, and multi-stage contextual retrieval served via FastAPI with real-time
inference.
• Fine-tuned large language models using LoRA and qLoRA for domain-specific use cases,
optimising model behaviour and retrieval quality for production performance; established model
versioning, experiment tracking, and reproducibility workflows throughout.
• Applied advanced prompt engineering techniques including chain-of-thought (CoT), few-shot,
and in-context learning to improve LLM output quality; evaluated performance using Ragas
style metrics (faithfulness, answer relevancy, context precision).
• Implemented LLM observability and tracing in production environments — structured logging,
monitoring dashboards, and API performance metrics — ensuring system reliability and enabling
rapid debugging.
• Engineered NLP and information retrieval pipelines processing 50, 000+ customer reviews,
applying text processing, embedding-based retrieval, and sentiment classification to improve
customer satisfaction insights by 20%.
• Productionized ML pipelines on AWS and Azure through CI/CD automation and monitoring,
reducing deployment time by 25% and maintaining high availability of inference services.
• Architected IoT predictive maintenance platforms integrating sensor telemetry with ML models
across edge hardware, profiling compute-constrained workloads and reducing equipment
downtime by 35%.
• Developed real-time vehicle number plate recognition systems using computer vision and
OpenCV pipelines, improving monitoring accuracy by 30%.
• Delivered predictive healthcare models for heart disease and diabetes achieving AUC up to 0.99
and precision of 95% using PyTorch deep learning architectures.
• Spearheaded Generative AI and speech deep learning research achieving 2.5% word error rate
in voice cloning and up to 100% accuracy in speech emotion recognition.
• Built and maintained ETL data pipelines to extract, transform, and load data from diverse
structured and unstructured sources into ML-ready formats for downstream model training and
evaluation.
• Optimised AdaBoost ensemble inference performance, improving model efficiency by 10%;
engineered unsupervised customer segmentation frameworks predicting purchasing behaviour
with 75% accuracy.
• Collaborated across engineering, product, data science, and business teams throughout the
SDLC — from discovery to deployment — translating complex AI capabilities into clear product
features and communicating results to non-technical stakeholders.
• Led cross-functional AI workshops with 10+ stakeholders to define enterprise AI requirements,
delivering a unified AI roadmap aligned with business strategy.

• Implemented ML and deep learning research algorithms in Python across NLP, signal
processing, and biomedical domains; developed breast cancer detection diagnostic models
achieving 96% accuracy using advanced deep learning architectures.
• Designed and ran controlled ML experiments with full tracking of hyperparameters, model
versions, and evaluation metrics; improved speech recognition accuracy from 75% to 85% using
DSP and deep learning techniques.
• Classified ECG signals using machine learning and signal processing pipelines, exceeding
baseline performance by 7% through structured experimentation and reproducible model
training workflows.
• Implemented predictive analytics-driven energy optimisation systems delivering 6% measurable
energy savings.
• Produced clear technical documentation for all models, pipelines, and experiments; contributed
to 2 published research outputs and won the Most Publications in Academic Year award.
• Mentored and supervised a 10-member research team, improving delivery timelines by 25%
and achieving quality scores of 4.2/5.
• Collaborated with product engineering teams to streamline model validation workflows,
reducing verification cycles by 50%. Python | LangChain LCEL FAISS Chroma FastAPI PostgreSQL Docker AWS
EC2 PyTorch Ragas Hardware