Lay Naik

• Architected & built end‑to‑end AI agent infrastructure
Solely designed and engineered a complex multi-agent enterprise framework using ideyaLabs' Quad‑Agent Architecture, coordinating DEV, INFRA, QA, and BA agents to deliver autonomous decision-making and deployment pipelines

• Developed a scalable “Dev Agent” platform
Created a custom agent from scratch to automate code generation at scale—handling requirements intake, task assignment, and production-grade code output, built for effectively unlimited horizontal scaling

• Deployed and fine‑tuned LLMs on-premise
Led hardware-level deployment of small transformer LLM models using on-prem GPU clusters—managing provisioning, latency optimization, model tuning, and deployment pipelines. Demonstrated full-stack expertise from model to infrastructure

• Technology & tools: Python · LangChain · Hugging Face · Prompt Engineering · Multi-agent orchestration · LLM fine‑tuning · GPU cluster deployment · CI/CD workflows · Scalable system design

Requirement Builder and Project Planner.

• Led the end-to-end development of AI backends for both products, architecting scalable systems that efficiently serve a large user base.
• Implemented advanced retrieval techniques, including Retrieval-Augmented Generation (RAG), contextual retrieval, two-stage retrieval, and hybrid search methodologies, enhancing information accuracy and relevance.
• Integrated multiple Large Language Models (LLMs) such as OpenAI, Anthropic, LLaMA, and Mistral, optimising prompts and system performance to meet diverse user needs.
• Deployed LLMs using Ollama, ensuring efficient hosting and scalability across platforms.
• Managed MongoDB databases for dynamic data retrieval and storage, supporting real-time user interactions.
• Developed AI observability tools and generated embeddings using models from OpenAI, NVIDIA, and Voyage, improving model transparency and performance monitoring.
• Utilised various vector databases to optimise embedding storage and retrieval processes, enhancing system responsiveness.
• Analysed and processed user requests to generate content aligned with the Software Development Life Cycle (SDLC), streamlining project planning and execution.
• Enhanced deployment processes by containerising applications with Docker, ensuring consistent and scalable environments across development, testing, and production stages.

Rocket.new – AI-driven code generation.

• Transitioned to the flagship product Rocket.new following the development of Requirement Builder and Project Planner, contributing to its inception and growth.
• Engineered agent-based systems for automated code generation, integrating LLMs through the Model Control Protocol (MCP) to streamline development workflows.
• Utilised various LLMs and auxiliary tools to optimise output quality and efficiency.
• Fine-tuned a compact LLM specifically designed for embedding generation tasks, improving embedding accuracy.
• Explored and implemented diverse prompting techniques to enhance result accuracy and relevance.
• Developed an intelligent agent capable of integrating code components and autonomously resolving bugs, thereby decreasing debugging time.

Leadership and mentorship.

• Oversaw the recruitment process for new AI team members, including reviewing application materials, evaluating technical tasks, and conducting interviews.
• Designed and implemented comprehensive training programmes for interns, ensuring alignment with job requirements, and continuous skill development.
• Managed a team of two interns, providing ongoing mentorship, performance evaluations, and professional development support, leading to their successful transition into full-time roles.

• As an AI Intern at DhiWise Pvt Ltd, I was responsible for leveraging advanced AI technologies to create innovative solutions, and enhance existing systems. My role involved a broad range of tasks within the AI domain, from model development, to system integration.
• Development of Recruitment System: Designed and implemented a state-of-the-art recruitment system to streamline the hiring process. Utilised machine learning algorithms to enhance candidate matching and prediction accuracy.
• RAG Pipeline for Code Generation: Developed and optimised a RAG (Retrieval-Augmented Generation) pipeline specifically for code generation processes. Improved the pipeline's efficiency and accuracy through iterative testing, and refinement.
• Vision Model Enhancement: Worked on enhancing the accuracy of existing object detection models using computer vision techniques. Applied advanced algorithms and conducted experiments to boost model performance and reliability.
• Generative AI: Worked on generative AI models to create innovative solutions for various business needs. Conducted experiments to enhance the generative capabilities and model efficiency.
• AI Integration and Deployment: Integrated AI models with existing systems, and ensured seamless deployment. Collaborated with cross-functional teams to ensure the AI solutions meet business requirements.
• JSON objects and JavaScript: Worked extensively with JSON objects to manage and manipulate data. Utilised JavaScript for scripting and automating tasks related to AI model integration.

• As an AI Intern at Logictrix Infotech, I was involved in the end-to-end development of a complex email classification system. My responsibilities spanned the entire project lifecycle, from database creation to deployment on AWS, providing me with comprehensive hands-on experience in AI system development.
• Development of Email Classification System: Designed and implemented a multi-model email classification system from scratch. Utilized various classification transformers to categorize emails accurately.
• Reinforcement Learning Implementation: Applied reinforcement learning techniques to enhance the performance and accuracy of the email classification system. Continuously improved the model through iterative training and feedback loops.
• Database Creation and Management: Created and managed a robust database to store and process email data. Ensured data integrity, security, and efficient retrieval for model training and evaluation.
• Deployment on AWS: Deployed the email classification system on Amazon Web Services (AWS) for scalable and reliable operation. Configured AWS services to support the application, including EC2, S3, and RDS.
• Comprehensive Project Lifecycle Management: Managed the entire project lifecycle, from initial planning and database creation to model development and final deployment. Collaborated with cross-functional teams to ensure seamless integration and operation of the system.