Samarth J.

Key Skills

Software

Other

Top Subject Matter

Image captioning / computer vision (Flickr8k, InceptionV3, LSTM)

Supervised ML for tabular fraud detection (SMOTE, XGBoost, AdaBoost)

Generative AI (text-to-music pipeline)

Top Data Types

Image

Text

Computer Code Programming

Top Task Types

Data Collection

Fine-tuning

Text Generation

Freelancer Overview

Image Caption Generator project (TensorFlow, InceptionV3, LSTM) — built and trained an image captioning model on Flickr8. Brings 1+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include TensorFlow, scikit-learn, and Other. Education includes Bachelor of Technology, Indian Institute of Technology (BHU) Varanasi (2026). AI-training focus includes data types such as Computer Code and Programming and labeling workflows including Data Collection, Fine-tuning, and Text Generation.

Labeling Experience

Image Caption Generator project (TensorFlow, InceptionV3, LSTM) — built and trained an image captioning model on Flickr8k.

Data Collection

Contributed to AI/ML model development by building a captioning pipeline trained on Flickr8k. The work involved preparing and using labeled image-caption pairs to learn mappings from visual features to text outputs. This contributed to improving caption fluency and relevance via supervised learning. • Used a pre-trained InceptionV3 CNN for feature extraction from images. • Trained an image captioning model combining CNN and LSTM networks with Keras. • Worked with the Flickr8k dataset, leveraging its image-caption annotations. • Evaluated improvements in generated caption quality (fluency and relevance).

2025 - 2025

AI Music Generator project (Next.js, FastAPI, Python, AWS S3) — text/lyrics-driven music generation with generative models.

OtherText Generation

Built an AI music generation system that produces original music from textual input using AI models. The project required preparing text/lyrics prompts and using model-driven generation to create outputs from those prompts. Credits-based backend services supported controlled usage of generation capabilities. • Implemented a full-stack SaaS for AI-driven music generation from text or lyrics. • Integrated multiple generative models (ACE-Step, Qwen2-7B, SDXL-Turbo) for production. • Developed a serverless FastAPI backend with Modal and Inngest. • Added credit-based payments and authentication for users generating music.

Not specified

Vehicle Insurance Fraud Detection project — ML classification with imbalance handling (SMOTE) and ensemble models.

Fine-tuning

Applied machine learning techniques to detect vehicle insurance fraud and address dataset imbalance for better predictive performance. The project included preprocessing and resampling of minority-class examples to improve classifier learning. Model evaluation focused on achieving strong F1 score, reflecting effective detection quality. • Performed exploratory data analysis (EDA) to understand feature patterns and correlations. • Implemented SMOTE to oversample the minority class in imbalanced data. • Trained ensemble models (e.g., XGBoost and AdaBoost) for classification. • Reported an F1 score of 0.9716 to validate precision/recall balance.

Not specified

Education

Work History