Lakshmi Priya Ramisetty

Data Engineer - GenAI

InTheLoop INC

Oct 2024 - Present

• Fine-tuned Vision-Language Models (VLMs), such as Qwen2-VL and Gemini 1.5 Pro, using Supervised Fine-Tuning (SFT) techniques to create a fabric damage detection system and integrated dynamic pricing strategies to adjust item pricing based on their damage levels.
• Optimized Qwen2-VL by modifying DeepSpeed’s pipeline to support Vision-Language Models (VLMs), improving inference speed by 2.5x.
• Accelerated ML model deployment and lifecycle management using Vertex AI to automate model training, fine-tuning, and deployment.
• Automated GCP infrastructure provisioning using Terraform to manage and deploy resources for data engineering and ML workflows.
• Designed a scalable pipeline using Apache Beam, Dataflow, to scrape and download over 125,000 images from the web, streamlining large-scale data collection with a microservices architecture
• Leveraged DINO v2 and KMeans clustering to group images based on feature embeddings, followed by background removal
• Utilized Gemini 1.5 Pro for good and bad labelling classification to enhance background removal accuracy, refining dataset quality through prompt-based fine-tuning

Data, Machine Learning Engineer Volunteer

PopStock Educational Services

Aug 2024 - Dec 2024

• Built a serverless RAG system on Google Cloud using BigQuery, Gemini, and Cloud Functions to deliver real-time, context-aware stock market insights while reducing operational costs by 60% through efficient resource utilization and serverless architecture.
• Launched a simulated stock market platform pilot program, successfully introducing it to 5 schools in New York to enhance financial literacy and hands-on learning for students.
• Automated JSON-based textual script generation for animations using Gemini 1.5 Pro, enabling seamless integration with Unreal Engine.
• Collected and preprocessed over 100 million records of historical stock data for Fortune 100 companies, ensuring data integrity through advanced feature engineering and normalization techniques.
• Designed and deployed a complex multi-layer LSTM model, improving predictive accuracy by 30% through detailed hyperparameter tuning and regularization.

AI/ML Engineer Intern

FinServ Experts

May 2024 - Aug 2024

• Fine-tuned state-of-the-art Large Language Models (LLMs) like Llama2 and Mistral using parameter-efficient techniques such as LoRA and QLoRA, achieving highly accurate and contextually relevant responses for financial applications.
• Created custom, domain-specific datasets tailored to financial use cases, including investment analysis, risk assessment, and regulatory compliance, optimizing model performance and reducing false positive rates by 35% in critical financial workflows.

Data Engineer

Egen

Mar 2021 - Jul 2023

• Developed several ETL pipelines leveraging Dataflow, Pub/Sub, Dataproc and BigQuery, processing and harmonizing large datasets 50% faster than the industry standard.
• Led the development of a comprehensive end-to-end healthcare demonstration using 7+ GCP tools, including BigQuery, BigQuery ML, Vertex AI, and Dialogflow, showcased at Google’s headquarters, illustrating advanced data analytics and AI capabilities.
• Migrated 8 dashboards from Tableau to Looker, creating LookML models from scratch that significantly enhanced data visualization.
• Established standards for hosting, CI/CD pipelines, Docker containerization, and Kubernetes orchestration. Implemented secure networking with GCP VPC, subnets, firewall rules, and IAM role-based access controls to enhance efficiency and security
• Achieved a 70% reduction in pipeline execution time by implementing a decision tree harmonization strategy, optimizing data processing workflows.
• Ingested and processed 10 TB of healthcare data comprising over 400 million records in BigQuery for high-volume analysis.
• Optimized SQL query performance on large datasets using database partitioning, reducing execution time from 10 minutes to 2 minutes.
• Utilized Google Cloud Workflows to orchestrate complex ETL pipelines, incorporating robust error handling and automated retry mechanisms to ensure reliability and achieving an impressive average uptime of 99.9%.

Machine Learning Engineer Intern

Accenture

Jan 2020 - Jul 2020

• Automated document processing using Tesseract OCR, reducing processing time to 2 seconds per document and enhancing data extraction accuracy by 30%.
• Conducted A/B tests to evaluate the effectiveness of different OCR algorithms, resulting in a 25% improvement in data extraction accuracy and efficiency.
• Researched reinforcement learning techniques, including Q-Learning, DQN, and PPO, for Collaborative Robots (CoBots) to enhance adaptive behavior and interaction, supported by a comprehensive literature review and detailed software bot taxonomy.

Languages & Databases

Libraries & Frameworks

Cloud Tools

Big Data, ETL & APIs

Tools