TryCrucible — Prove your AI engineering skills

Scenario

You are given a summarisation model and a dataset of 100 documents with human-written reference summaries. The model output contains 5 planted failure modes. Design an eval suite that measures quality across multiple dimensions and identifies where the model fails.

Expected output

Eval suite catches all 5 planted failure modes with < 5% false positives

Dataset files

How the three files work together

manifest.json

Lists all available files and the evaluation contract. Read this first to understand the challenge structure.

dataset.json

The full source data (FAQ entries, documents, issues, etc.) your solution is built on. Index it, embed it, or process it.

test_inputs.json

10–20 test queries the evaluator will run against your solution at submission time. Your code reads this file and writes results.json.

📦

Dataset files

Scoring rubric

Clarity5%

Eval results are interpretable and actionable

LLM usage20%

LLM-as-judge prompts are tight, consistent, and calibrated against human labels

Robustness10%

Evals are deterministic or have low variance across runs

Correctness25%

Detects all planted failure modes, false positive rate < 5%

Architecture20%

Eval dimensions are orthogonal, well-defined, and independently scorable

Language-free evaluation

Build your solution in any language or framework — Python, TypeScript, Go, Rust, Java, C#, or anything else. The dataset artifacts may be in one language; your implementation does not need to match. TryCrucible evaluates the behaviour of your system, the quality of your AI workflow, your verification strategy, and the reproducibility of your submission — not your language choice.

Submission requirements

A public GitHub repository link
A Dockerfile in the repo root — any language or framework; the evaluator builds and runs your container
Your solution reads test_inputs.json from /workspace/test_inputs.json (use the TEST_INPUTS_PATH env var) and writes results.json to /workspace/results.json (use the RESULTS_PATH env var)
A decisions.md — 3–5 sentences on the key architectural and AI-workflow choices you made
The system must be fully reproducible — we clone, build, and run it against real test inputs

Evaluation contract

When you submit, the evaluator runs these steps in order:

1Clone your public GitHub repository
2Build your container from the Dockerfile in the repo root
3Mount test_inputs.json at /workspace/test_inputs.json (TEST_INPUTS_PATH env var)
4Run your solution in a network-isolated sandbox (5 min limit, 512 MB RAM)
5Read results.json from /workspace/results.json (RESULTS_PATH env var)
6Score correctness against hidden ground truth, then score architecture, AI workflow, robustness, and clarity

// I/O contract — same schema for every challenge

Input (provided by evaluator)

// test_inputs.json
[
  { "id": "t1", "input": { ... } },
  { "id": "t2", "input": { ... } }
]

Output (written by your solution)

Create a free account to start

Already have one? Sign in

Write evals for a summarisation model