AI Benchmark Researcher (Remote)

Joblet-AI; India; Remote; INR 3,240,000 - 7,776,000; Full time

Job summary

Joblet-AI is seeking an analytical researcher and red-teaming specialist to craft evidence-based benchmark tasks that reveal weaknesses in SOTA AI models. You will uncover hard-to-find facts, create precise prompts, and test them against leading LLMs, diagnosing failure modes in search and reasoning paths.

Your work shapes AI evaluation frameworks to push frontier models toward rigorous, real-world scenarios, with a contractor engagement of up to 40 hours per week and overlap with PST.

Qualifications

• 1+ years in investigative research, fact-checking, AI benchmark task authoring, red-teaming, or technical QA review.
• Proven experience authoring, testing, or reviewing tasks for AI benchmarks such as GAIA, BrowseComp, SimpleQA, SWE-bench, Terminal-Bench, or Humanity’s Last Exam (HLE).
• Intermediate-to-advanced OSINT skills; ability to navigate archives, records, or obscure databases.
• Strong analytical capability to inspect model reasoning chains, search logs, and source selection to diagnose failures.
• Clear, unambiguous benchmark prompts with primary-sourced answer keys.

Responsibilities

• Obscure Fact Discovery: research niche topics to identify factual truths not surfaced by direct searches.
• Source Verification: verify facts against primary sources, document source URLs.
• Task Construction: draft clear benchmark tasks to expose structural weaknesses in model research strategies.
• Red-Teaming & Model Testing: run tasks against designated SOTA models to trigger failures.
• Failure Diagnosis & Debugging: trace logs to pinpoint where and why the model failed.
• Pattern Scaling & Review: apply research patterns across domains and review tasks by peers.

Skills

Investigative research

OSINT skills

Model debugging

Technical writing

Benchmark testing

Education

Bachelor's degree in relevant engineering field

Tools

GAIA

BrowseComp

SimpleQA

SWE-bench

Job description

• Role Type: Full-time
• Location: Remote

Role Overview

We are seeking an analytical researcher and red-teaming specialist to build evidence-based web-research benchmark tasks that reliably expose weaknesses in state-of-the-art (SOTA) AI models (e.g., Claude Opus 5 Max). In this role, you will discover hard-to-surface facts, build precise research tasks around them, test them directly against leading LLMs, and diagnose the exact failure mechanisms in the model's search and reasoning paths.

Your work will directly shape AI evaluation frameworks, ensuring next-generation frontier models are tested against rigorous, non-trivial real-world research scenarios.

Job Requirements :

1. Core Experience: 1+ years in investigative research, fact-checking, AI benchmark task authoring, red-teaming, or technical QA review.

2. Benchmark Familiarity: Proven experience authoring, testing, or reviewing tasks for AI benchmarks such as GAIA, BrowseComp, SimpleQA, SWE-bench, Terminal-Bench, or Humanity’s Last Exam (HLE).

3. Analytical & OSINT Skills: Intermediate-to-advanced web search and Open Source Intelligence (OSINT) skills; ability to navigate academic archives, local records, or obscure databases.

4. Model Trajectory Debugging: Strong analytical capability to inspect model reasoning chains, search logs, and source selection to diagnose root causes of failure.

5. Precision Writing: Ability to author clear, unambiguous benchmark prompts with bulletproof, primary-sourced answer keys.

Job Responsibilities:

1. Obscure Fact Discovery: Deeply research niche topics to identify factual ground truths that cannot be surfaced via simple, direct search queries.

2. Source Verification: Rigorously verify facts against primary or highly credible sources, cross-check for contradictory evidence, and document exact source URLs.

3. Task Construction: Draft clear, answerable benchmark tasks designed specifically to expose structural weaknesses in model research strategies.

4. Red-Teaming & Model Testing: Execute tasks against designated SOTA models to confirm they reliably trigger intended model failures.

5. Failure Diagnosis & Debugging: Trace model run logs, search trajectories, and intermediate outputs to pinpoint exactly where and why the model failed (distinguishing genuine model limitations from ambiguous prompts).

6. Pattern Scaling & Review: Apply validated "model-breaking" research patterns across new domains and, for strong performers, review tasks authored by peer trainers.

Strong Plus (Preferred Qualifications)

7. Bachelor or higher degree from relevant engineering domain is a must for this position.

8. Prior experience specifically designing prompt-injection or reasoning-trap tasks that break SOTA models.

9. Advanced OSINT capabilities or published investigative/academic research demonstrating deep source verification.

10. Niche domain expertise in areas rich with hard-to-find records (e.g., local government filings, regional sports history, legal archives, historical patents).

11. Experience serving as a lead reviewer, benchmark auditor, or quality owner on AI evaluation projects.

1. Commitments Required: Upto 40 hours per week with 4 hours of overlap with PST.

2. Engagement type: Contractor

3. Engagement Length: upto 4 weeks

Evaluation Process

1. Candidates will be require to submit a Job Interest Form & assessment.

2. Talents finalized in assessment will go through assessment, review & will proceed further accordingly.

What You’ll Do

1. Obscure Fact Discovery: Deeply research niche topics to identify factual ground truths that cannot be surfaced via simple, direct search queries.

2. Source Verification: Rigorously verify facts against primary or highly credible sources, cross-check for contradictory evidence, and document exact source URLs.

3. Task Construction: Draft clear, answerable benchmark tasks designed specifically to expose structural weaknesses in model research strategies.

4. Red-Teaming & Model Testing: Execute tasks against designated SOTA models to confirm they reliably trigger intended model failures.

5. Failure Diagnosis & Debugging: Trace model run logs, search trajectories, and intermediate outputs to pinpoint exactly where and why the model failed (distinguishing genuine model limitations from ambiguous prompts).

6. Pattern Scaling & Review: Apply validated "model-breaking" research patterns across new domains and, for strong performers, review tasks authored by peer trainers.