About Me

Building Intelligent, Scalable AI Infrastructure

I am an entry-level Machine Learning Engineer specializing in Computer Vision, Document AI, and Multimodal AI systems. I have a passion for design and implementation of high-throughput, data-intensive pipelines that process, validate, and extract structured intelligence from complex unstructured sources.

My experience spans training transformer-based vision models, orchestrating multi-stage data pipelines with Apache Airflow and Vertex AI, fine-tuning large language models (LLMs) via LoRA/PEFT, and building optimized, production-ready inference endpoints.

8.4B.Tech AI & DS CGPA
1IEEE Publication
1 YrML Internship Exp.
95%CV Model Accuracy

02 / Technical Skills

Core Competencies & Toolchain

01

Machine Learning & Core

PyTorchPythonSQLNumPyPandasScikit-learnTransfer LearningData Pipelines
02

Computer Vision & Document AI

Swin TransformerYOLO (Object Detection)Image ClassificationVertex AIOCR WorkflowsOpenCVData Augmentation
03

NLP & Large Language Models

LLM Fine-TuningLoRA / PEFTRAG PipelinesGemini APILLaMAConversational AIText Classification
04

Model Optimization & Systems

Mixed Precision Training (AMP)ONNX Export & Inference4-bit QuantizationOneCycleLRAdamWGradient ClippingApache Airflow
05

Backend, Cloud & Dev

FastAPIFlaskREST APIsDockerCloudflare WorkersGitPower BIExcel

03 / Professional Experience

Journey So Far

Machine Learning Intern

Intrust Innovation Labs
Feb 2025 – Feb 2026
  • Developed large-scale Document AI pipelines using Google Cloud Vertex AI to process scanned business documents and extract structured information from unstructured PDFs.
  • Built OCR-driven document processing workflows encompassing document ingestion, preprocessing, validation, deduplication, quality assurance, and structured output generation.
  • Designed automated data curation pipelines, enhancing document extraction accuracy and significantly reducing manual processing overhead.
  • Developed transformer-based computer vision models using PyTorch, optimizing classification accuracy through transfer learning, advanced data augmentation, and hyperparameter search.
  • Integrated deep learning models into production inference workflows via FastAPI/REST APIs and backend services.
  • Orchestrated multi-stage machine learning workflows using Apache Airflow to automate training data generation and model deployment cycles.

04 / Selected Projects

Innovative AI Implementations

IEEE PublishedLive

FarmSage Multimodal Platform

End-to-End AI Platform for Precision Agriculture

A comprehensive agricultural decision-support platform combining deep-learning computer vision models with real-time generative conversational intelligence.

PyTorchYOLOSwin TransformerFastAPIGemini APIRAGONNXVertex AIDockerCloudflare Workers
Key Achievements & Architecture
  • Developed a multilingual crop-disease detection and advisor system, combining a Gemini API RAG pipeline with live local weather forecasts.
  • Designed an end-to-end vision system using YOLO for object localization and Swin Transformer for fine-grained leaf disease classification, achieving 95% test accuracy.
  • Implemented robust training optimizations including Mixed Precision (AMP), OneCycleLR scheduling, AdamW optimization, gradient clipping, and advanced augmentations (MixUp, Random Erasing).
  • Optimized model serving latency by 30% by exporting weights to ONNX runtimes served via FastAPI and Cloudflare Workers.
GitHub

Aura Conversational Assistant

Quantized LLM Fine-Tuning & Deployment Workflow

An optimized conversational agent trained on customized domain data using advanced parameter-efficient fine-tuning (PEFT) and low-bit quantization.

LLaMA 3.2BLoRAPEFTFlaskgTTSHuggingFaceQuantization
Key Achievements & Architecture
  • Fine-tuned LLaMA 3.2B with 4-bit precision quantization and LoRA adapters, building an end-to-end model training, evaluation, and logging pipeline.
  • Designed dataset preparation, tokenization, model validation, and prompt-refinement scripts, reducing memory footprints during training.
  • Developed a lightweight Flask REST API supporting real-time text-to-speech and speech-to-text response features with low latency.
  • Managed end-to-end production pipelines including model pruning, serialization, and Docker containment.

05 / Research & Publications

IEEE Published Research

IEEE XplorePublished 2025

"Farm Sage: Multilingual Agricultural Q&A Assistant and Plant Leaf Disease Detection"

Presented at the 2025 10th International Conference on Smart Structures and Systems (ICSSS)
Chennai, India, 2025 · pp. 1–6

This research paper presents an end-to-end multimodal AI platform that combines object localization (YOLO) and fine-grained classification (Swin Transformer) to identify plant leaf diseases with 95% accuracy, integrated with a weather-aware, multilingual Retrieval-Augmented Generation (RAG) assistant for crop care.

Academic Education

Education Background

Bachelor of Technology (B.Tech)

Artificial Intelligence and Data Science

2022 – 2026

K.S. Rangasamy College of Technology

Tiruchengode, Tamil Nadu, India

8.4 / 10Cumulative GPA
Core Focus Areas

Machine Learning, Data Structures & Algorithms, Deep Learning, Statistical Analysis, Database Management Systems.