Apex
User: The sole HSE officer at an Azerbaijani manufacturing SME (20–250 employees). These firms can’t afford an HSE department, so one person is responsible for every SƏTƏM obligation across the whole facility. Problem: Risk assessments, work permits, briefing records, and instructions are written by hand in Word, one at a time. It’s slow and inconsistent, and clauses get missed or cite outdated rules. When documentation is missing or wrong, the firm faces state inspections, fines, suspended operations, and personal liability for the officer if an incident happens. The only alternatives are hiring expensive consultants or doing it late and partially.
Preliminary score
- Value for the user19.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI16.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing5.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility7.5 / 15
Data requirements, running costs and a clear next step.
- Originality5.5 / 10
A meaningful difference from common solutions.
What the judges wrote
- Claude Opus 5.5Preliminary51/100
- Strengths
- Your user is sharply defined: the sole HSE officer at an Azerbaijani manufacturing SME who carries every SƏTƏM obligation alone, with real consequences such as inspections, fines and personal liability. The core flow is concrete. A user picks a document type, describes the work in plain language or uses a sample scenario, gets a draft grounded in RAG over public SƏTƏM legislation, and exports a PDF. Your stack (Supabase with row-level security and pgvector, Noto Sans for Azerbaijani characters in jsPDF) shows you thought about multi-tenant data and local-language rendering.
- To improve
- Your testing covered only PDF rendering. Test the AI output itself: check a set of generated risk assessments and permits against the legislation, count missed or outdated clauses, and show cases where retrieval cited the wrong article. Compare against the current way of working, for example how long the officer takes to write a permit in Word versus with your tool, and how many required clauses each version includes. Add a measurable outcome, per-document API and embedding costs, and a concrete next step such as a pilot with a named SME or review by an HSE consultant. A README or short demo video would also let judges verify the end-to-end flow.
- GPT-6 AstraPreliminary56/100
- Strengths
- Your focus on the sole HSE officer at an Azerbaijani manufacturing SME identifies a specific user with a concrete documentation burden. Your described workflow gives AI a substantive role by combining facility details with retrieval over local SƏTƏM legislation to draft required documents. Your report of corrupted PDFs and overlapping sections identifies a genuine usability failure rather than relying only on generic success claims.
- To improve
- Your evidence would be stronger with one complete example showing facility input, retrieved clauses, generated text and a successfully exported PDF. Your testing should verify legal citations and missing safety requirements with an HSE reviewer, document PDF regression results, and compare drafting time and correction rates against Word. Your claim that the AI text was fine needs supporting checks, with qualified human approval before safety-critical documents are used. Your feasibility plan should quantify cost per document, explain legislation updates and facility-data handling, and define a small pilot as the next step.
The team
Farid GuluzadeLead
developer
Nadir Nadirli
business
Gunel Ali-zade
product