Java Developer- Ai Training

Talantir; Deutschland; Remote; EUR 55.000 - 96.000; Vollzeit

Zusammenfassung

Talantir is seeking a highly skilled independent contractor to join our AI evaluation team. The role focuses on evaluating and improving how AI systems reason about code, generate programming solutions, and explain technical concepts across complexity levels.

You'll rigorously assess AI-generated code responses, perform fact-checking, run code to validate outputs, and annotate results with actionable feedback to improve model correctness and reliability.

Qualifikationen

• BS/MS/PhD in Computer Science or related field.
• Significant real-world software engineering experience.
• Expertise with at least one major programming language.

Aufgaben

• Evaluate LLM-generated responses for coding accuracy and reasoning.
• Conduct fact-checking using trusted sources and references.
• Execute code to validate outputs with appropriate tools.
• Annotate model responses with strengths and improvements.
• Assess code quality, readability, and algorithmic soundness.
• Ensure responses follow system guidelines and consistent evaluation taxonomies.
• Provide reproducible evaluation artifacts and actionable feedback.

Kenntnisse

Software engineering

Open-source contributor

LLM evaluation experience

Ausbildung

BS/MS/PhD in Computer Science

Tools

Python

Java

C++

JavaScript

Go

Rust

Jobbeschreibung

Work Mode: Remote

Engagement Type: Independent Contractor

Schedule: Full-Time or Part-Time Contract

Language Requirement: Fluent English

Role Overview

We partner with leading AI teams to improve the quality, usefulness, and reliability of general-purpose conversational AI systems.

This project focuses specifically on evaluating and improving how AI systems reason about code, generate programming solutions, and explain technical concepts across various complexity levels.

The role involves rigorous technical evaluation of AI-generated responses in coding and software engineering contexts.

What You’ll Do

• Evaluate LLM-generated responses to coding and software engineering queries for accuracy, reasoning, clarity, and completeness
• Conduct fact-checking using trusted public sources and authoritative references
• Conduct accuracy testing by executing code and validating outputs using appropriate tools
• Annotate model responses by identifying strengths, areas of improvement, and factual or conceptual inaccuracies
• Assess code quality, readability, algorithmic soundness, and explanation quality
• Ensure model responses align with expected conversational behavior and system guidelines
• Apply consistent evaluation standards by following clear taxonomies, benchmarks, and detailed evaluation guidelines

Who You Are

• You hold a BS, MS, or PhD in Computer Science or a closely related field
• You have significant real-world experience in software engineering or related technical roles
• You are an expert in at least one relevant programming language (e.g., Python, Java, C++, JavaScript, Go, Rust)
• You are able to solve HackerRank or LeetCode Medium and Hard–level problems independently
• You have experience contributing to well-known open-source projects, including merged pull requests
• You have significant experience using LLMs while coding and understand their strengths and failure modes
• You have strong attention to detail and are comfortable evaluating complex technical reasoning, identifying subtle bugs or logical flaws

Nice-to-Have Specialties

• Prior experience with RLHF, model evaluation, or data annotation work
• Track record in competitive programming
• Experience reviewing code in production environments
• Familiarity with multiple programming paradigms or ecosystems
• Experience explaining complex technical concepts to non-expert audiences

What Success Looks Like

You identify incorrect logic, inefficiencies, edge cases, or misleading explanations in model-generated code, technical concepts, and system design discussions

Your feedback improves the correctness, robustness, and clarity of AI coding outputs

You deliver reproducible evaluation artifacts that strengthen model performance