PMI Sleep 5 · Krilltech
A credit analysis workspace bringing together customers, operations and evidence to support human review. 1st place in IBM’s independent technical evaluation, held after the hackathon. Demonstration with simulated data.
Product concept and business rules, Product Model Canvas and pitch script, with contributions to implementation alongside the team.
Product Model Canvas · React · TypeScript · Python
1st place in IBM’s independent technical evaluation
Our Sleep 5 team placed first in the technical evaluation of the Challenge with IBM Bob, with Krilltech Inteligência de Crédito. The Hackathon PMI-DF organizers shared the review after the event ended.
This result is separate from the hackathon ranking. The evaluation considered only the submitted GitHub repositories, using its own criteria. The pitch and Project Canvas were not considered.
The review highlighted the project for its technical and scientific rigor: automated domain and UI tests, scoring with explicit mathematical foundations, and transparent documentation of implemented and simulated functionality.
Context and solution
Krill Tech’s challenge was to anticipate defaults, judicial reorganization and insolvency in agriculture. Our solution brings customers, operations and evidence into a credit-analysis workspace with human review.
The evaluation identified a Python scoring tool based on log-odds, a watsonx Orchestrate integration through the @tool decorator, and a React/TypeScript frontend with seven areas. It also recognized unit and E2E tests, parity between Python and frontend rules, and distinctions between types of environmental violations.
My contribution
I contributed to the product concept, business rules, Product Model Canvas, pitch script and implementation alongside the team. The technical achievement belongs to the team; I do not claim sole authorship of every module or test praised in the review.
Connectivity limitations during the event prevented some contributions from being recorded under my account. Some work was integrated through teammates’ accounts, so commit authorship does not fully represent my participation.
Recognized limitations
This is a demonstration, not a production backend. The review notes scripted chat responses without a connected LLM, no live legal-data collection or production API, and synthetic CSV data. The watsonx Orchestrate integration does not mean the entire solution was integrated in production.
Evaluation evidence
Original Portuguese images supplied by the Hackathon PMI-DF organizers. The podium reports 8.06, while the individual review reports 8.10. Because the source material differs, this portfolio highlights the placement without treating either overall score as definitive.
