Manvith Kumar Ullal | AI Enthusiast

Scroll to Explore

SKILLS

PROJECTS

EXPERIENCE

Full Stack Developer Intern

May 2026 – July 2026

MyClickBook

Contributed to the development of MyClickBook, a SaaS-based appointment booking and business management platform for salons and service providers. Worked across frontend and backend modules, implementing authentication, onboarding workflows, API integrations, localization, testing, and UI improvements while collaborating in an Agile team using GitLab.

Key Contributions & Highlights:

Developed and enhanced React Native features for the Vendor application, translating Figma designs into responsive mobile interfaces.
Built complete onboarding workflows, including role selection, phone authentication, OTP verification, business setup, staff onboarding, portfolio management, weekly schedules, and goal selection.
Implemented Google Sign-In, authentication flows, user profile updates, and resolved login and username synchronization issues.
Integrated and tested REST APIs for authentication, vendor management, onboarding, inventory, and business workflows.
Implemented application localization (i18n) and updated localization test cases to support multilingual functionality.
Developed vendor profile management features, including portfolio galleries, salon services, staff management, and inventory UI.
Completed 35+ GitLab issues and feature requests across multiple modules during the internship.

Tech Stack

React NativeTypeScriptExpoREST APIsGitLabFigmaFirebaseGoogle OAuthi18n

Gen AI & Cloud Computing Intern

June 2026 – July 2026

IBM SkillsBuild & BharatCares

Successfully completed the AICTE IBM SkillsBuild Gen AI & Cloud Computing Internship in collaboration with IBM SkillsBuild and BharatCares.

Key Contributions & Highlights:

Gained practical knowledge of cloud computing fundamentals, cloud service and deployment models, virtualization, Docker containerization, and APIs.
Developed hands-on experience with containerized application deployment, cloud-based software development, and cloud databases.
Explored cloud security, identity and access management (IAM), database security, and cloud application deployment workflows through practical simulations.
Strengthened understanding of modern enterprise cloud technologies while enhancing problem-solving skills in real-world cloud environments.

Tech Stack

Cloud ComputingIBM CloudDockerAPIsIAMCloud DeploymentVirtualizationDatabase SecurityCloud DatabasesContainerization

AI/ML Intern

April 2026 – May 2026

BrandandBrandz

Worked on the development of an AI-powered Receptionist Automation System using n8n, Groq AI, Airtable, REST APIs, and webhook integrations.

Key Contributions & Highlights:

Contributed to designing and implementing automated appointment booking workflows, including scheduling, rescheduling, cancellations, and reminder systems.
Integrated APIs and managed databases using Airtable, configured webhooks, and debugged API authentication and execution issues.
Collaborated on building conversational AI workflows, enhancing automation efficiency and user interaction handling.
Improved overall system reliability using structured development practices.

Tech Stack

n8nGroq AIAirtableCal.com APIREST APIsPostmanWebhooksAutomation WorkflowsAI Agents

RESEARCH

Multimodal Deep Learning for Feature-Level Fusion in Automated Arecanut Quality Classification Using RGB and X-ray Imaging

Manvith Kumar Ullal, Shivasubrahmanya KC, Preran Rai, G. Nisha Bangera, Vidya Kolur, Duddela Sai Prashanth, Rakshith Bhandary, Priya Kamath

Under ReviewSSRN Preprint • 2026

Designed a novel multimodal deep learning framework for automated arecanut quality classification by combining RGB surface images with X-ray imaging, enabling simultaneous analysis of external appearance and internal structural defects. The study systematically evaluated four multimodal learning paradigms and demonstrated the effectiveness of feature-level fusion for agricultural quality assessment.

Key Highlights

Developed a multimodal AI pipeline integrating 11,123 RGB and 10,076 X-ray images for non-destructive quality grading.
Evaluated four deep learning strategies: feature-level fusion, domain adaptation, multimodal learning, and ensemble learning.
Achieved 99.80% classification accuracy and 99.81% Macro F1-score using an independent encoder feature-fusion architecture.
Performed statistical validation using Stratified 5-Fold Cross Validation, paired t-tests, bootstrap testing, and Wilcoxon signed-rank tests.

TensorFlowPyTorchMobileNetV3EfficientNetV2ConvNeXtSwin TransformerOpenCVCUDAPython

View Paper

An End-to-End Deep Learning Framework for Handwritten Tulu Lipi Recognition and AI-Assisted Script Learning

Manvith Kumar Ullal, Arshith, Preran Rai, Thushar, Rajeshwari R Shettigar

Under ReviewResearch Paper • 2026

Proposed an AI-powered educational platform for preserving the Tulu language through handwritten character recognition and interactive script learning. The system combines a CNN-based recognition model with a Flask inference API and a React-based learning platform to provide real-time handwriting evaluation and learner feedback.

Key Highlights

Expanded a handwritten Tulu dataset from 9,959 to approximately 30,000 images using advanced data augmentation techniques.
Designed and trained a custom CNN-based handwriting recognition model integrated with a Flask REST API for real-time inference.
Built an AI-assisted learning platform featuring handwriting evaluation, pronunciation support, learner progress tracking, and interactive feedback.
Developed a modular architecture using React, Node.js, Flask, and TensorFlow for scalable deployment and real-time predictions.

PythonTensorFlowKerasFlaskReactNode.jsOpenCVNumPyPillowCNN

An Explainable Federated Learning Framework for Privacy-Preserving Pneumonia Detection Using Chest X-ray Images

Arshith, Manvith Kumar Ullal, Preran Rai, Thushar, Rajeshwari R Shettigar

Under ReviewIEEE Research Paper • 2026

Proposed an explainable federated learning framework for privacy-preserving pneumonia detection from chest X-ray images. The framework combines an SE-enhanced DenseNet121 model with FedAvg and FedProx across three simulated hospital clients, while Grad-CAM provides visual explanations of model predictions and cross-dataset evaluation measures generalization.

Key Highlights

Developed an SE-enhanced DenseNet121 model for privacy-preserving pneumonia detection using federated learning.
Compared FedAvg and FedProx across three simulated hospital clients using the RSNA Pneumonia Detection Challenge dataset.
Achieved 82.49% accuracy and 88.52% AUC with FedAvg, while FedProx achieved 82.16% accuracy and 88.63% AUC with higher recall of 77.27%.
Evaluated cross-domain generalization on an independent Labeled Chest X-ray Images dataset, where FedProx achieved 70.51% accuracy, 54.62% recall, and 90.35% AUC.
Applied Grad-CAM to provide visual explanations and verify whether the models focused on clinically relevant pulmonary regions.