Hire Dakshitha - Fine-Tuning Engineer

* Led design and deployment of a RAG-based sales assistant using LangChain, GPT-4o, and Pinecone, processing 1TB+ of * CRM data and reducing query resolution time by 57% (4.2mins to 1.8 mins) across 1,200+ users. * Developed and deployed a lead scoring model using XGBoost and AWS SageMaker Feature Store, serving 12M+ monthly * predictions under 110 ms p99 latency; increased MQL-to-SQL conversion rate by 14%. * Designed and implemented an end-to-end MLOps pipeline using MLflow, Apache Airflow, Docker, and Kubernetes (EKS); * reduced model retraining time from 3 days to 6 hours and decreased deployment failures by 38%. * Built a real-time churn prediction system using Apache Kafka and Spark Streaming, processing 400 GB/day of behavioral * data; enabled proactive retention strategies, reducing annual churn by $300K. * Fine-tuned Llama 3.1 (8B) using QLoRA (parameter-efficient fine-tuning) on 120K support tickets; reduced escalation rates by * 31% and saved $90K annually in Tier-1 support costs. * Implemented model monitoring and data drift detection using Evidently AI, reducing silent model performance degradation * incidents by 60%. * Collaborated with Product and Data Engineering to define ML feature contracts, SLAs, and production readiness standards

* Developed and deployed fraud detection models using Random Forest and LightGBM with Bayesian hyperparameter * optimization, preventing $1M in annual fraud losses and achieving 91% precision. * Built cloud-native ML solutions on AWS, leveraging SageMaker, S3, AWS Glue, and AWS Lambda for scalable training, data * pipelines, and RESTful API deployment; reduced infrastructure costs by 30% while maintaining 99.8% SLA compliance. * Designed and optimized deep learning models (CNNs, RNNs) using PyTorch and TensorFlow for computer vision and timeseries forecasting, improving model accuracy by 15%, reducing false positives by 10%, and increasing inference speed by 20%. * Engineered scalable ML pipelines using Apache Spark, MLflow, and Apache Airflow, reducing model retraining time by 18% * and improving pipeline reliability and monitoring. * Automated CI/CD pipelines for machine learning models using GitHub Actions and Docker, supporting 12+ production models * and reducing release cycles from 2 weeks to 4 days with full audit compliance. * Deployed machine learning models across classification, regression, and NLP tasks using Python, Scikit-learn, and * TensorFlow, improving overall model performance by 22%.