Prabhu B.

Key Skills

Software

Other

Internal/Proprietary Tooling

Top Subject Matter

Large Language Model (LLM) response evaluation and RLHF preference annotation

Software code review and AI-generated code correctness evaluation (Python, Java, JavaScript)

Enterprise software quality assurance and AI-assisted test automation evaluation

Top Data Types

Text

Computer Code Programming

Document

Top Task Types

RLHF

Object Detection

Fine-tuning

Red Teaming

Evaluation/Rating

Computer Programming/Coding

Function Calling

Prompt + Response Writing (SFT)

Text Generation

Freelancer Overview

My experience in AI training and data labeling spans three professional roles running concurrently at points, which means the volume of work adds up fast. As an LLM Trainer with Outlier AI, I completed 500+ structured preference annotation tasks across coding, math, creative writing, and general knowledge. The work covered RLHF preference ranking, adversarial prompt engineering, and line-by-line code evaluation in Python and JavaScript, checking AI outputs for factual accuracy, logical coherence, instruction-following, and safety. I kept strong inter-annotator agreement scores throughout and got genuinely good at catching hallucinations, failure modes, and edge cases before they slip through the cracks. Alongside that, my current role as an Analyst at Deloitte keeps that evaluation muscle active daily. I assess AI-generated code in production environments, integrate LLM tooling into QA workflows, and build test coverage that accounts for the weird, boundary-pushing cases most people miss. What I bring that tends to stand out is the combination of technical depth and evaluative instinct. I understand why a model produces a given output, not just whether it looks right on the surface, and that makes my feedback specific and actually useful for training iterations. Whether it is ranking LLM responses, reviewing AI-written code, classifying NLP outputs, or auditing datasets for quality and bias, I have done it in real professional settings with real accountability attached.

Labeling Experience

Analyst (Deloitte India, Chennai, On-site)

Other

Performed AI-assisted testing evaluations by assessing AI-generated test scripts for correctness, coverage completeness, and edge-case handling. Integrated LLM tooling into QA workflows to automate repetitive evaluation tasks and support prompt quality review. Conducted systematic code review and logic validation across Java and JavaScript to ensure reliability of evaluated outputs. • Designed and maintained automated test suites using Selenium and Playwright for UI logic, API behavior, and data integrity checks. • Evaluated AI-generated test scripts for correctness, coverage, and edge-case handling. • Integrated LLM tooling into QA workflows for repetitive evaluation automation. • Produced documentation covering test strategies, defect analysis, and AI evaluation findings for reproducibility.

2025 - Present

LLM Trainer & AI Evaluator (Outlier AI, Remote, Part-time)

OtherTextRLHF

Trained and evaluated LLMs by ranking generated responses and performing preference annotation using structured RLHF methodologies. Assessed factual accuracy, logical coherence, instruction-following, and safety alignment to support fine-tuning and model alignment. Conducted adversarial prompt testing and code correctness reviews to improve training feedback quality. • Completed 500+ preference annotation tasks across coding, mathematics, creative writing, and general knowledge. • Engineered adversarial prompts to surface hallucinations and instruction-following failures. • Reviewed AI-generated Python and JavaScript for logical correctness, security vulnerabilities, and edge-case handling. • Maintained high inter-annotator agreement and annotation consistency across batches.

2024 - 2025

Education

Work History