 Engineered a multi-agent AI system using JAX and TensorFlow to generate and validate biomedical hypotheses, integrating RAG pipelines over

40M+ genomic and chemical entries, reducing experimental search time from 48 to 31 hours per batch.

 Designed and trained transformer-based models for protein interaction prediction and drug repurposing using supervised and unsupervised

learning approaches, achieving a predictive accuracy increase of 22% on benchmark datasets.

 Developed a cloud-native architecture on Google Cloud Platform (GCP) leveraging Kubernetes for scalable model training and inference,

reducing end-to-end model deployment time by 40% and enabling continuous experimentation.

 Integrated SQL pipelines to efficiently preprocess and query multi-modal biomedical datasets, enabling rapid retrieval of structured and

unstructured data for downstream AI reasoning.

 Operationalized NLP-driven biomedical chatbots using transformer architectures and LangChain frameworks to provide researchers with

interactive, context-aware insights, accelerating literature search and experimental planning by 30%.

 Formulated and fine-tuned transformer models using Hugging Face for biomedical text understanding, enabling automated literature

summarization and hypothesis extraction, reducing research effort by 40% while maintaining 95% semantic accuracy across 10K+ articles.

 Conducted end-to-end R&D on multi-agent reasoning systems that combine supervised learning, unsupervised clustering, and reinforcement

learning to simulate complex biological interactions, contributing to three peer-reviewed publications in computational biology.

 Optimized AI model performance and resource utilization on GCP by leveraging distributed training, mixed-precision computation, and realtime monitoring dashboards, improving inference efficiency by 25% while maintaining model accuracy on large-scale biomedical datasets.

 Programmed and deployed predictive models for 5G network traffic and customer churn using Python, PyTorch, and ensemble methods

(Random Forest, XGBoost, LightGBM), achieving 92% accuracy and reducing network downtime forecasting errors by 28%.

 Devised and fine-tuned BERT-based NLP pipelines for analyzing telecom customer interactions, leveraging LLM embeddings to classify service

requests, improving automated ticket resolution by 35% while reducing manual triage workload.

 Implemented end-to-end CI/CD pipelines using Docker and AWS SageMaker for scalable model training and deployment, reducing model

release cycle time from 14 days to 5 days across multiple environments.

 Built real-time analytics pipelines on AWS Cloud integrating MongoDB and time-series forecasting models to predict network load and optimize

5G bandwidth allocation, processing 2M+ traffic events daily, improving network utilization by 18%.

 Performed hyperparameter optimization using Optuna and Ray Tune for deep learning and ensemble models, boosting predictive model F1

scores by 12% and ensuring robustness across diverse telecom datasets

 Established Responsible AI practices, including bias detection, fairness evaluation, and model explainability in production pipelines, ensuring

compliance with regulatory standards while maintaining high model performance across telecom and biomedical datasets.

 Architected comprehensive model validation and unit testing frameworks to ensure production reliability, including statistical performance

monitoring, drift detection, and reproducibility of results in cloud deployments.

 Collaborated with cross-functional teams to integrate ML models with telecom APIs, enabling personalized service recommendations and

automated analytics dashboards, driving a 22% increase in customer engagement and revenue optimization.