Ishika Jain

Key Skills

Software

Roboflow

Top Subject Matter

Computer Vision/Workplace Safety & Access Control

Conversational AI/Agentic Systems

Healthcare Document AI

Top Data Types

Image

Text

Top Task Types

Object Detection

Fine-tuning

Classification

Freelancer Overview

AI-ML Trainee - Data Annotation & Model Training. Brings 4+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Roboflow, Internal, and Proprietary Tooling. Education includes Bachelor of Technology, ABES Institute Of Technology (2025). AI-training focus includes data types such as Image and Text and labeling workflows including Object Detection and Fine-tuning.

Labeling Experience

AI/ML Engineer - LLM Fine-tuning & Training Data Curation

TextFine Tuning

Fine-tuned LLMs (Qwen2.5-7B-Instruct) using structured summary extraction from transcripts to enhance downstream automation. Focused on preparing and curating text-based datasets for model tuning, ensuring data suitability and relevance. Performed LLM training iterations to reach desired automation accuracy improvements. • Created training data by extracting summaries from dialogue transcripts. • Evaluated and adjusted model performance based on labeled outputs. • Leveraged internal/proprietary tooling for dataset management during LLM fine-tuning. • Supported AI/ML pipeline accuracy enhancements with direct data curation efforts.

Fine-tuned LLMs (Qwen2.5-7B-Instruct) using structured summary extraction from transcripts to enhance downstream automation. Focused on preparing and curating text-based datasets for model tuning, ensuring data suitability and relevance. Performed LLM training iterations to reach desired automation accuracy improvements. • Created training data by extracting summaries from dialogue transcripts. • Evaluated and adjusted model performance based on labeled outputs. • Leveraged internal/proprietary tooling for dataset management during LLM fine-tuning. • Supported AI/ML pipeline accuracy enhancements with direct data curation efforts.

2025 - Present

End-to-End Conversational Humanoid AI Avatar System (Video-to-Video Generation)

VideoSegmentation

Led the R&D and engineering of a fully in-house, end-to-end conversational humanoid AI avatar system. The project focused on transitioning from third-party SDKs to a proprietary pipeline featuring real-time video-to-video generation trained on human performance datasets. Key tasks included: Data Curation: Developing datasets for custom multilingual TTS, specifically focusing on Hindi Devanagari to Roman alignment for semantic speech synthesis. Model Training: Fine-tuning multimodal frameworks to ensure high controllability and synchronization between audio and video outputs. Optimization: Implementing emotion-aware dialogue systems and optimized audio pipelines to reduce latency by 65% and distortion by 99%. Quality Measures: Adhered to strict contextual consistency standards, supporting 1k+ concurrent sessions with a 98% improvement in responsiveness compared to previous iterations.

Led the R&D and engineering of a fully in-house, end-to-end conversational humanoid AI avatar system. The project focused on transitioning from third-party SDKs to a proprietary pipeline featuring real-time video-to-video generation trained on human performance datasets. Key tasks included: Data Curation: Developing datasets for custom multilingual TTS, specifically focusing on Hindi Devanagari to Roman alignment for semantic speech synthesis. Model Training: Fine-tuning multimodal frameworks to ensure high controllability and synchronization between audio and video outputs. Optimization: Implementing emotion-aware dialogue systems and optimized audio pipelines to reduce latency by 65% and distortion by 99%. Quality Measures: Adhered to strict contextual consistency standards, supporting 1k+ concurrent sessions with a 98% improvement in responsiveness compared to previous iterations.

2025 - 2026

Data Scientist Intern – Generative AI (Healthcare Data Labeling & Fine-tuning)

TextFine Tuning

Implemented LoRA-based fine-tuning of vision-language models (VLMs) and LLMs on proprietary healthcare data. Created and managed training datasets and performed targeted document understanding tasks to improve extraction accuracy. Directly enhanced model outcomes, achieving 98–99% document understanding accuracy. • Prepared and labeled ground-truth datasets for model training and evaluation. • Collaborated on evaluating model benchmarks using standardized labeled data. • Ensured data privacy and compliance while managing sensitive healthcare information. • Utilized internal tools for annotation and pipeline integration in a clinical AI setting.

Implemented LoRA-based fine-tuning of vision-language models (VLMs) and LLMs on proprietary healthcare data. Created and managed training datasets and performed targeted document understanding tasks to improve extraction accuracy. Directly enhanced model outcomes, achieving 98–99% document understanding accuracy. • Prepared and labeled ground-truth datasets for model training and evaluation. • Collaborated on evaluating model benchmarks using standardized labeled data. • Ensured data privacy and compliance while managing sensitive healthcare information. • Utilized internal tools for annotation and pipeline integration in a clinical AI setting.

2025 - 2025

AI-ML Trainee - Data Annotation & Model Training

RoboflowImageObject Detection

Annotated over 5500 images using Roboflow for multiple computer vision projects in a training role. Applied augmentation techniques to enhance model robustness and fine-tuned YOLOv7 for specific object detection tasks. Contributed directly to dataset preparation for AI model training and improved detection accuracy to 95%. • Labeled images for PPE kit detection, multi-person face recognition, and parking sticker recognition. • Used bounding boxes and other annotation methods as required by project needs. • Ensured high data quality and accuracy throughout the annotation process. • Supported deployment of models trained on the annotated data.

Annotated over 5500 images using Roboflow for multiple computer vision projects in a training role. Applied augmentation techniques to enhance model robustness and fine-tuned YOLOv7 for specific object detection tasks. Contributed directly to dataset preparation for AI model training and improved detection accuracy to 95%. • Labeled images for PPE kit detection, multi-person face recognition, and parking sticker recognition. • Used bounding boxes and other annotation methods as required by project needs. • Ensured high data quality and accuracy throughout the annotation process. • Supported deployment of models trained on the annotated data.

2023 - 2024

Education

Work History