The project collection
03 / Handshake AI · Project Dynamo

AI evaluation tasks

Building clear specifications and verifiable programming tasks to evaluate AI-generated solutions.

Evaluation workflow illustration · Public overview of the task-design process
Evaluation workflow illustration · Public overview of the task-design process03 / 06
My role

AI Trainer · Contract

Timeline

May — August 2026

Focus

Python / C · Docker · Linux · Automated testing

01

Turning a problem into a specification

For Project Dynamo at Handshake AI, I designed programming tasks with clear requirements and an objective way to evaluate the resulting solution. The work required careful attention to the boundary between an intended challenge and an ambiguous instruction.

02

Building the verification

I worked with reference solutions and automated checks using Python, C, Docker, and Linux. Task design drew on dependency analysis, workflow recovery, and performance testing, connecting the written specification to observable behavior.

03

What this work developed

The experience strengthened my ability to reason about failure cases, explain expected behavior, and create repeatable verification. This case study describes the process at a high level; internal task code and private evaluation materials remain confidential.

SpecifyClear requirements
ImplementReference solutions
VerifyAutomated checks
Continue exploring

See the work in context.

About Handshake AI
Up next / Meticulous Health Solutions

Better care operations

Turning administrative and compliance requirements into usable workflows and clear documentation.