About me

Hi, I’m Yash Nayi. I’m a Data Scientist and Generative AI practitioner who’s always been curious about how intelligence can grow from simple ideas. That curiosity eventually turned into my work as a data scientist and generative AI practitioner. What fascinates me most is how data turns into decisions and how those decisions ripple into real-world impact.

My journey started with a question: how can data tell a story? That question led me to Udacity’s Data Scientist Nanodegree, where I learned to clean, analyze, and model data using Python, SQL, Apache Spark, and Scikit-learn. I built projects that turned messy datasets into clear insights and visual stories. I still remember how exciting it felt to see patterns appear from what first looked like chaos.

As AI evolved, so did my curiosity. During Udacity’s Generative AI Nanodegree with Accenture, I fine-tuned LLMs, built RAG-based chatbots, and experimented with tools like Stable Diffusion and LangChain. That’s when I stopped just prompting models and started building with them. I began to see that AI could reason and assist, not just generate.

Now at CREWASIS.AI, I focus on building multi-agent AI systems: agents that can collaborate, reason, and act together. Using LangChain, LangGraph, and N8N, I design agentic workflows that mimic real teamwork. Each agent has its own role and decision-making ability, and together they solve problems that go beyond what a single model could do. It’s been exciting and sometimes chaotic to see how close we can get to systems that actually think through problems instead of just answering them.

Looking back, my path has always been about curiosity and persistence. I’ve broken more prototypes than I can count, but each one taught me something new about how machines and people learn. I care about building AI that people can understand, trust, and use, something that feels more like a teammate than a tool.

I’m still learning every day, experimenting, and sharing what I find. If you’re exploring the same space, I’d love to connect and keep pushing the boundaries of what intelligent systems can really do.

Skill & Expertise

  • design icon

    Programming Languages & Web Development

    Python SQL JavaScript R Dart Java PHP HTML/CSS Bootstrap(CSS)

  • design icon

    Tools

    Jupyter Notebook Git Git version control Google Colab Cursor Cursor AI Jira Notion Slack Discord Microsoft Suits Google Suits Anaconda R studio VS Code PyCharm Android Studio

  • design icon

    Machine Learning & Deep Learning

    Sci-Kit Learn PyTorch TensorFlow XGBoost LightGBM Regression Classification Superviesd ML Unsupervised ML Neural Network ClearML (Tool)

  • design icon

    GenAI

    Embedding (Vector) LLM Fine-Tuning Transfer Learning Benchmark Retrieval Augmented Generation Hugging Face PEFT (LoRA, QLoRA, Inference) Hyper-parameter Tuning N8N - AI Agents Workflow LangGraph LangChain LangSmith Streamlit - AI Agent Platform Ollama

  • design icon

    Data Analysis & Visualization Tools

    Pandas Numpy SciPy Matplotlib Plotly(Interactive Visualization) Seaborn Power BI Tableau

  • design icon

    Big Data & Distributed Computing

    AWS Apache Spark Hadoop MongoDB SQL JSON

Current Project!

  • N8N

    YouTube Auto Transcript, Document & Note Generation | N8N AI Workflow

    Built an end-to-end automation pipeline in N8N that extracts YouTube video transcripts, converts them into structured documents and summary notes, and organizes them in a Google Drive workspace. Integrated OpenAI APIs for topic segmentation, summarization, and key insight generation, enabling automated content analysis and note creation for long-form videos. This system reduced manual documentation time by over 90%, making it ideal for researchers, content creators, and students managing large video datasets.


    N8N OpenAI API YouTube Data API YouTube Video Link Google Drive API Markdown & PDF automation


  • N8N

    Automated Job Application & Resume Matching | N8N + AI Agents

    Developed a multi-step job application automation using N8N workflows that fetches AI/ML Engineer roles from LinkedIn, Greenhouse, and Ashby APIs, parses job details, and performs AI-based resume-to-job matching. Used LLM-based skill extraction and semantic similarity scoring to calculate a “match score” between job requirements and resume content. Automatically stores key details — including Job ID, Title, Company, Match Score, and Recruiter Email — into a Google Sheet, enabling faster, data-driven job targeting.


    N8N OpenAI API RSS Google Sheet API Google Drive API Google Gemini API LinkedIn API Greenhouse API Ashby API JSON Automation


  • Gemma

    Lightweight Fine-tuneing using PEFT (PEFT Gemma 2B)

    This project leverages Parameter-Efficient Fine-Tuning (PEFT) techniques to optimize large-scale language models, significantly reducing training time and resource usage while maintaining high performance. By selectively adjusting key parameters using methods like adapter modules and low-rank factorization, the project demonstrates a scalable and efficient approach for enhancing model accuracy across various NLP tasks.


    PEFT LoRA QLoRA Bitsandbytes Transformers Tokenization Google Colab Gemma 2B HuggingFace


    GitHub Github huggingface HuggingFace
  • Daniel lewis

    Retrieval-Augmented Generation (RAG) for a Custom Financial Chatbot using GenAI

    This project focuses on building a domain-specific chatbot using the Retrieval-Augmented Generation (RAG) technique, aimed at delivering high-quality answers to financial queries. The system integrates OpenAI’s GPT models with a curated dataset of financial documents to enhance the quality of responses using contextually relevant data.


    OpenAI GPT-3.5 Turbo Query Embedding Semantic Retrieval Context-Aware Generation Text-embedding-3-small Cosine similarity Python pandas requests scipy.spatial.distance Data Manipulation Data Analysis Batch Processing


    GitHub Github Custom AI Chatbot
Daniel lewis
quote icon

Retrieval-Augmented Generation (RAG) for a Custom Financial Chatbot using GenAI

This project focuses on building a domain-specific chatbot using the Retrieval-Augmented Generation (RAG) technique, aimed at delivering high-quality answers to financial queries. The system integrates OpenAI’s GPT models with a curated dataset of financial documents to enhance the quality of responses using contextually relevant data.

Daniel lewis
quote icon

Lightweight Fine-tuneing using PEFT (PEFT Gemma 2B)

This project leverages Parameter-Efficient Fine-Tuning (PEFT) techniques to optimize large-scale language models, significantly reducing training time and resource usage while maintaining high performance. By selectively adjusting key parameters using methods like adapter modules and low-rank factorization, the project demonstrates a scalable and efficient approach for enhancing model accuracy across various NLP tasks.

Tools & Libraries I used for my projects!

Resume

Experience

  1. Data Engineer & Ml-Ops Intern

    CREWASIS.AI | Aug 2025 — Present

    At CREWASIS.AI, I design and deploy multi-agent AI systems that think, reason, and collaborate like real teams. I build dynamic LangGraph workflows that integrate models such as Claude 3.5 Sonnet, Opus 4.1, DeepSeek R1, and Llama 3 70B, enabling role-based reasoning for virtual personas like data scientists, product managers, and engineers. I lead the development of a conversational AI platform featuring dynamic model switching, persona simulation, and long-term memory persistence — allowing AI agents to adapt their tone, strategy, and responses based on evolving context. To ensure reliability, I architect an evaluation framework that combines A/B testing, white-box, and black-box analysis to benchmark reasoning accuracy, consistency, and user satisfaction. I also develop JSON-based observability and feedback loops to capture user interactions and model outputs, driving continuous optimization of agent behavior and prompt performance. By orchestrating scalable multi-agent workflows with LangChain, LangGraph, LangSmith, and N8N AI agents, I help transform static LLMs into dynamic, autonomous systems that achieve over 30% higher task accuracy and contextual adaptability. (In essence: I empower AI agents to reason, adapt, and improve like humans — at production scale.)

    AWS LLM Models RAG N8N - AI Agents Workflow LangChain LangGraph LangSmith Streamlit API integration AI Agent Testing A/B Testing Prompt Engineering Prompt Testing Prompt Optimization & Evaluation

  2. Flutter Mobile applications development Intern

    BrainyBeam Technologies Pvt. Ltd. | Jan 2022 — Dec 2022

    When tasked with modernizing event coordination, I built a cross-platform mobile app from scratch using Flutter (Dart) and a Python REST API. I collaborated with backend developers to streamline data flow between server and client using Flask and Postman testing, ensuring seamless performance under load. To enhance collaboration, I introduced Git version control and modular architecture, improving maintainability and team velocity by 25%. Ultimately, the app became a central tool for managing large events efficiently and intuitively

    Flutter Dart SQL REST API Python Cloud(AWS)

  3. Python Developer

    BrainyBeam Technologies Pvt. Ltd. | Fed 2021 — Aug 2021

    Faced with redundant business processes, I developed data-driven web applications in Django and SQL to automate workflows that previously required manual intervention. I optimized ETL queries and backend logic, reducing data latency by 40% and improving page load speeds. I also implemented standardized testing and CI/CD practices, raising code reliability and reducing production bugs. These improvements elevated the team’s delivery standards and accelerated deployment cycles.

    Python REST API SQL Postman (API Testing) Django Flask MangoDB Database Management VS code PyCharm

  4. Software Developer Intern

    Unity Infoway | Jan 2019 — July 2019

    As a junior developer entering the Agile world, I built responsive Java-based web interfaces using Eclipse, XML, and Bootstrap. I integrated APIs and SQL databases to deliver dynamic, data-rich web experiences, improving user engagement. Working within Agile sprint cycles, I learned to collaborate across design and backend teams, contributing to faster feature delivery and an understanding of end-to-end product development.

    Java SQL RESET API HTML CSS JavaScript JQuery XML

Volunteer Work Experience

  1. Generative AI Engineer - University Of New Haven

    Nov 2024 — Feb 2025

    Digital Twin Research GenAI Deep Learning Machine Learing Data Analysis Hyper-parameter Tuning

Education

  1. University Of New Haven, Connecticut, United States

    2023 — 2025

    Master in Data Science

  2. Government of Engineering College (Government Technological University), Gujarat, India

    2019 — 2022

    Bachelor of Science in Information Technology

  3. Government Polytechnic (Government Technological University), Gujarat, India

    2016 — 2019

    Diploma in Information Technology

Certification

  1. Building LLM Applications With Prompt Engineering

    Apr 2025

    from NVIDIA

    Certificate – Building LLM Applications With Prompt Engineering

  2. Generative AI

    Apr 2025

    Generative AI Nanodegree from Udacity

    Certificate – Building LLM Applications With Prompt Engineering

  3. Data Scientist

    Sep 2024

    Data Scientist Nanodegree from Udacity, built in collaboration with IBM, earned in September 2024.

    Certificate – Building LLM Applications With Prompt Engineering

Projects

  • SocioLens – LLaMA-3.2-3B Interactive Chat Interface

    Built a dual-theme conversational web app using HTML, CSS, and JavaScript that connects to the SocioLens LLaMA-3.2-3B model, enabling users to analyze public health policies through an AI-driven chat interface with real-time assistance and contextual awareness.

    Capstone Group Project
    README.md

  • NdLinear – Ensemble AI for Vision

    Built and evaluated a parameter-efficient CNN layer (NdLinear) on CIFAR-10 and Fashion-MNIST, reducing model parameters by over 80% while slightly improving accuracy and maintaining comparable speed and memory efficiency.

    Ensemble AI - Project
    README.md

  • OlfecNet – Intelligence

    Designed a neural network model to predict odor characteristics from molecular and GC-MS data, solving the challenge of mapping chemical structures to sensory perception with high accuracy.

    University Project
    README.md

  • AI Photo Editing with SAM and Diffusion Inpainting

    Created an interactive AI-powered image editor using Meta’s Segment Anything Model (SAM) and Diffusion Inpainting to let users modify image backgrounds or subjects via text prompts, solving the challenge of intuitive, high-quality photo manipulation without manual editing.

    Certification Project
    README.md

  • HomeMatch – AI-Powered Real Estate Recommendation System

    Developed an AI-based real estate agent using NLP, vector search, and LLMs to match user preferences with property listings, solving the challenge of finding personalized home recommendations through conversational search.

    Certification Project
    README.md

  • Lightweight Fine-Tuning to a Foundation Model

    Implemented a streamlined fine-tuning pipeline using Hugging Face Transformers to efficiently adapt pre-trained language models to custom datasets, improving task-specific performance with minimal computational cost.

    Certification Project
    README.md

  • Retrieval-Augmented Generation (RAG) Financial Chatbot

    Built a domain-specific financial chatbot using OpenAI’s API and vector embeddings to retrieve relevant financial data and generate accurate, context-rich responses to user queries.

    Certification Project
    README.md

  • Classification of Handwritten Digits Using an MLP

    Built a Multi-Layer Perceptron (MLP) neural network to classify handwritten digits from the MNIST dataset, achieving high accuracy in digit recognition.

    Solo Project
    README.md

  • IPL Matches Inning Analysis

    Analyzed IPL match data to identify key performance metrics and trends, providing insights into team strategies and player performances using PowerBI visualizations.

    University Group Project
    README.md

  • Semantic-Segmentation-using-DeepLabV3

    Implemented DeepLabV3 for semantic segmentation to accurately classify and segment objects in images, solving pixel-level classification challenges in computer vision.

    University Group Project
    README.md

  • NLP-Biomedical-NER

    Developed a Named Entity Recognition system for biomedical text to extract and classify medical entities, addressing information extraction challenges in healthcare documents.

    University Group Project
    README.md

  • Boston AirBnb Listing Analysis

    Analyzed Boston AirBnb listings to identify pricing patterns and factors affecting rental success, helping hosts optimize their listings for better performance.

    Certification Project
    README.md

  • ML Disaster Response

    Built a machine learning pipeline to classify disaster messages for emergency response teams, enabling faster routing of critical messages to appropriate agencies.

    Certification Project
    README.md

  • Recommendations with IBM

    Developed a recommendation system using collaborative filtering to suggest articles to users, solving content discovery challenges in knowledge platforms.

    Certification Project
    README.md

  • Starbucks Capstone

    Analyzed Starbucks customer behavior and promotion effectiveness to optimize marketing strategies and improve customer engagement through data-driven insights.

    Certification Project
    README.md

  • Starbucks Promotion Optimization

    Optimized Starbucks promotion strategies using machine learning to predict customer response, maximizing ROI and improving targeted marketing campaigns.

    Certification Project
    README.md

  • Shiv Shakti Furniture

    Built a responsive e-commerce website for furniture retail, solving online presence and customer accessibility challenges for a local business.

    Solo Project
    README.md

Contact

Contact Form