Hi, I'm Ramya Lakshmi! 🙋🏻‍♀️

AI Software Engineer

University of Florida · Jan 2026 – Present

• Shipped an Academic Advisor Agent serving 1,000+ students, dynamically planning multi-semester schedules based on career goals.
• Achieved 100% real-time schedule accuracy and eliminated data staleness by building a Model Context Protocol (MCP) server that wrapped the UF REST API into dynamic agentic tools.
• Built a hybrid retrieval pipeline using a Nemotron model across 100K+ records, combining Text-to-SQL for relational queries with GraphQL for hierarchical graph traversal, reducing query failures by 95%.
• Reduced LLM inference cost and latency by 40% through prompt and context engineering, separating system prompt from runtime context and optimizing retrieved evidence and tool outputs by managing context windows for grounded responses.
• Boosted AI agent reliability by 80% using an automated evaluation pipeline with AWS Bedrock and CloudWatch observability, continuously tracking response correctness, faithfulness, latency, and cost.
• Load-tested and monitored production reliability under concurrent traffic, driving 70% satisfaction across 200+ users.

Machine Learning Researcher

Trustworthy Engineered Autonomy Lab · May 2025 – Present

• Developed TRACE, a training-free, black-box detector for prompt-consistency hallucinations in text-to-video generation by reconstructing prompt tokens from content-aware keyframes with a frozen VLM.
• Built a CI/CD pipeline automating a zero-shot VLM scoring system to prevent prompt-consistency hallucinations, cutting the evaluation timeline by 60%.
• Engineered an image processing pipeline for object detection and segmentation with YOLO-World and SAM2 models, achieving approximately 90% accuracy on physically distorted content.

AI Engineer

Yes Holidays · May 2025 – Aug 2025

• Deployed a customer support chatbot with a React front-end and FastAPI back-end, integrating an end-to-end RAG pipeline to handle flight, hotel, and holiday package inquiries.
• Designed a hybrid-search RAG architecture using LlamaIndex and parent-child chunking for accurate retrieval, coupled with a deterministic intent router that escalates complex queries to humans via WhatsApp.
• Achieved <2s average response latency using the Qdrant vector database and an OpenAI mini-LLM, automating 80% of routine customer questions and reducing manual ticket volume while monitoring execution traces with LangSmith.

AI/ML Engineer

Vernekar Associates Pvt. Ltd. · Jan 2023 – Jul 2024

• Achieved 23% energy savings and won "Best New Product of the Year" by developing Lumina Sync, a web app that combines real-time solar data and ArcGIS spatial analysis with Linear Regression, XGBoost, LSTM, and GRU models to deliver month-wise lighting recommendations.
• Reduced manual architectural drawing-review time by 75% by building an ML-based QA system that reviewed standardized door, window, stair, and lift details across plan, section, and elevation views.
• Mastered generative AI tools and subsequently trained fellow architects and new engineers on them, accelerating overall project progress by 20% and fostering a culture of innovation.

Architect

Siraj & Renu Architects & Interior Design · Aug 2022 – Dec 2022

• Developed detailed construction documentation using AutoCAD, translating design requirements into build-ready technical drawings.
• Collaborated with company leadership to create presentation assets for international clients using PowerPoint and Adobe Photoshop.
• Engineered residential floor plan concepts that balanced client requirements, spatial efficiency, and architectural constraints.

AR/VR Design Intern

Future Research Design Company · Dec 2021 – May 2022

• Awarded "Best Intern" for LiDAR-powered AR furniture visualizations delivering personalized 3D designs that secured a high-budget project.
• Collaborated with strategy and design teams to conduct market research for a branded showroom concept recognized with a Best Design Award.
• Developed standardized material palettes for the company's first branded showroom, establishing a reusable design framework for future locations.