Praveen Mishra

Key Skills

Software

Other

Roboflow

Labelbox

Don't disclose

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Computer Code Programming

Image

Video

Top Label Types

Prompt Response Writing SFT

Bounding Box

Segmentation

Freelancer Overview

I am an experienced AI Trainer and Data Annotation Specialist with a strong background in delivering high-quality training data for computer vision, NLP, and machine learning projects. My hands-on experience spans video, image, audio, and SQL-based data labeling, including fine-grained temporal annotation for action recognition models, prompt-response video labeling for social media content, and complex SQL command generation for LLM training. I have worked on platforms like Appen, Innodata, Atlus AI, Outliner, and Agents Only, consistently meeting demanding QA standards. My technical expertise includes Python, OpenCV, YOLOv8, TensorFlow, PyTorch, and SQL, and I have engineered solutions that optimize model accuracy and throughput for real-time detection systems. I am passionate about contributing precise, reliable data to power the next generation of AI models.

Labeling Experience

OPEN TRAIN AI

LabelboxImageQuestion AnsweringPrompt Response Writing SFT

Executed end-to-end quality assurance and validation of multimodal Visual Question Answering (VQA) datasets by auditing image-text QA pairs, identifying annotation discrepancies, and implementing corrective labeling to improve dataset integrity and model training reliability. • Applied structured taxonomy frameworks (Valid / Invalid / Incorrect / Cannot Judge), performed evidence-based content verification, and enforced linguistic, spatial, and factual accuracy standards to enhance data quality, reduce labeling errors, and optimize downstream AI model performance.

Executed end-to-end quality assurance and validation of multimodal Visual Question Answering (VQA) datasets by auditing image-text QA pairs, identifying annotation discrepancies, and implementing corrective labeling to improve dataset integrity and model training reliability. • Applied structured taxonomy frameworks (Valid / Invalid / Incorrect / Cannot Judge), performed evidence-based content verification, and enforced linguistic, spatial, and factual accuracy standards to enhance data quality, reduce labeling errors, and optimize downstream AI model performance.

2025

Agents Only

Don T DiscloseAudioAudio Recording

I have recorded high quality 2 person communication for a project. the role defines to record communication between 2 person with specific topic.

I have recorded high quality 2 person communication for a project. the role defines to record communication between 2 person with specific topic.

2025 - 2025

Outliner SQL LLM Training

OtherComputer Code ProgrammingFine TuningEvaluation Rating

Acted as a Machine Learning SQL Trainer, writing complex SQL commands to train LLMs in selecting optimal tools for complex data tasks. Designed Questions that require multiple Tool selection and validated SQL data for correct tool selection. Maintained accuracy for the project.

Acted as a Machine Learning SQL Trainer, writing complex SQL commands to train LLMs in selecting optimal tools for complex data tasks. Designed Questions that require multiple Tool selection and validated SQL data for correct tool selection. Maintained accuracy for the project.

2025 - 2025

Atlus Capture AI

OtherVideoSegmentationText Generation

Executed fine-grained temporal annotation (atomic labeling) for video data with timestamps, highlighting tasks at the second level for action recognition models. Helped team to enhance quality accuracy with other labelers by conducting regular Quality enhancement sessions

Executed fine-grained temporal annotation (atomic labeling) for video data with timestamps, highlighting tasks at the second level for action recognition models. Helped team to enhance quality accuracy with other labelers by conducting regular Quality enhancement sessions

2025 - 2025

Garbage Detection

RoboflowImageBounding BoxPoint Key Point

The goal of the project is to create a boundery box for human garbage sets into the image, which is highly compact and tight with the data frames.

The goal of the project is to create a boundery box for human garbage sets into the image, which is highly compact and tight with the data frames.

2025 - 2025

Show more

Education

Work History