You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$15,000 left to reach the funding goal.
I am Daniel Yoon Sik Choi, a University of Toronto LLM thesis researcher supervised by Professor Abdi Aidid. I want to build a small, open toolkit for testing whether people actually verify AI-assisted legal work. A human sign-off is not evidence that a review was effective. The toolkit will make the checks behind that sign-off observable.
Goals and approach
Over four months from funding availability, I will develop six fictional legal tasks with supplied sources, documented errors and independent scoring rubrics. Reviewers will decide which claims are supported, correct mistakes and escalate where information is missing. The interface will record checks, decisions, confidence and time. Expert walkthroughs will validate the materials; a small participant feasibility pilot will follow only after required institutional ethics approval. This is a methods-development project, not a powered claim about the effectiveness of oversight across all legal work.
How I will know it worked
The release will include six usable source packets, an error taxonomy, scoring guidance, a lightweight recording interface and a methods note. Expert review will identify ambiguity and assess whether the scoring captures substantive errors and harmful corrections. Any participant findings will be reported with their limitations. If ethics or recruitment prevents a pilot, the release will clearly identify expert walkthroughs and technical validation as such. Original materials and code will be openly licensed where rights permit; participant data will not be released without appropriate consent and protections.
Budget
Minimum US$5,000: $2,500 researcher time for task construction and documentation; $2,000 expert validation; $500 API and hosting. Target US$15,000: $8,000 researcher time and implementation; $3,000 independent expert review; $2,000 participant compensation subject to approvals; $2,000 API, hosting, reproducibility and release costs. The minimum scope is an expert-reviewed toolkit; the target adds a feasibility pilot and a fuller implementation. Funding is for the open research output, not customer acquisition or general Juridical AI runway.
Team and track record
I am applying as the individual researcher responsible for this project. My background includes a JD, an MA in philosophy, New York capital-markets practice, legal AI implementation and professional education. I have developed a thesis proposal on structured verification of AI-assisted transactional work and am developing Juridical AI (https://juridicalai.ca/). I do not claim completed participant studies or established product deployments.
Risks and failure
The main risks are tasks that are too ambiguous or easy, expert-scoring disagreement, implementation delays and approval or recruitment delays. The project is deliberately small. Independent review and transparent documentation will expose weaknesses. Even if a participant pilot is infeasible, an openly documented task set and account of its limitations would be useful.
Funding history and overlap
I have not yet reconciled my complete grant-receipt history for the last 12 months, so this proposal does not assert that I have received zero funding. Prior records report BlueDot Career Transition and Seed applications and a US$10,000 Transformative AI Fund individual-development application. An overlapping sao Fund toolkit application has been sent, and a US$7,500 BlueDot Rapid Grant request has now been submitted. These are applications, not representations of awards received. I will disclose actual awards, reconcile overlapping budgets and reduce or withdraw this request where another funder covers the same costs before accepting funds. Research support requests also include annotation and API credits, which will be accounted for separately.
AI assistance disclosure
AI assisted in organizing and drafting this proposal from my research plans and background. The work is proposed; statements about implementation, validation and participant research should be read accordingly.