Shubham N.

Key Skills

Software

Appen

Clickworker

CrowdSource

Labelbox

OneForma

Remotasks

Sama

Scale AI

SuperAnnotate

Telus

Don't disclose

Top Subject Matter

Visual multimodal AI content training and QA

Adversarial LLM evaluation and multilingual prompt-based training

LLM output evaluation

Top Data Types

Audio

Computer Code Programming

Text

Top Task Types

Computer Programming/Coding

Data Collection

Prompt + Response Writing (SFT)

RLHF

Text Summarization

Fine-tuning

Red Teaming

Transcription

Entity (NER) Classification

Evaluation/Rating

Text Generation

Object Detection

Classification

Question Answering

Freelancer Overview

Abaka.AI — Advanced AI Data Trainer (Visual Content). Brings 3+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Don't disclose, RaterHub, and CrowdCompute. Education includes Master's Degree in Commerce, Bundelkhand University (2021) and Professional Certificate in Data Science, Simplilearn (2022). AI-training focus includes data types such as Text, Audio, and Computer Code and labeling workflows including Fine-tuning, Red Teaming, and Evaluation.

Labeling Experience

Invisible Technologies — Advanced AI Data Trainer (QC & Contributor)

Text

Reviewed and rated Google Gemini outputs focused on truthfulness, fluency, and safety. Performed factuality and instruction-following checks across high-volume multi-turn conversations. Applied rater workflows to judge output quality for reliability and safety. • Evaluated truthfulness and fluency of model responses • Assessed safety and compliance signals • Checked instruction-following across multi-turn dialogs • Used RaterHub and CrowdCompute for review tasks

2025 - Present

DeepLLMData — Prompt Engineer

Don't discloseTextRed Teaming

Performed adversarial AI evaluation and multilingual LLM training using prompt-based test sets. Designed culturally grounded prompts to probe model reasoning, safety, and factual accuracy across Hindi, Hinglish, and English. Contributed to multilingual alignment-style training and evaluation. • Built and tested culturally grounded adversarial prompts • Evaluated reasoning, safety, and factual accuracy • Covered Hindi, Hinglish, and English language variants • Supported multilingual LLM training with prompt quality checks

2025 - Present

Abaka.AI — Advanced AI Data Trainer (Visual Content)

Don't discloseTextFine-tuning

Served as a generalist AI trainer for advanced multimodal visual-content systems. Drove development, evaluation, and improvement of models through annotation and reasoning tasks. Ensured outputs meet structured quality guidelines across training iterations. • Performed multimodal annotation and reasoning contributions • Supported evaluation loops for model improvement • Applied structured quality control procedures • Contributed to dataset/training quality for visual content systems

2025 - Present

Outlier AI — AI Data Trainer · Prompt Engineer · QC

OneFormaAudioTranscription

Trained Google Gemini with real-time speech and text data as part of AI training and evaluation workflows. Specialized in factuality QA, speech annotation, and prompt construction within RLHF and SFT loops. Annotated and reviewed bilingual outputs in Hindi and English to improve multilingual quality. • Performed speech annotation paired with text data • Conducted factuality QA for model outputs • Supported RLHF/SFT prompt construction and loop participation • Reviewed bilingual Hindi-English responses for quality

2024 - Present

AI Data Quality Reviewer - RWS Group

Scale AIAudioEntity (NER) Classification

Delivered factuality and audio reviewing for the Callisto project and supported quality assurance across annotation tasks. Ran review checks to ensure consistency of AI trainer outputs and alignment with required standards. Applied linguistic QA skills across speech and text evidence to validate response quality. • Factuality review for AI outputs in the project context • Audio and speech-related quality checks • Consistency checking across engagements • Support for multilingual or bilingual review workflows

2024 - 2025

Show more

Education

Work History