AI Enterprise SolutionsGPT & Claude

DaCy

BizSim is built for program managers at technoparks, university entrepreneurship centers and accelerators running cohorts of 50–250 participants. Today, they manage training, exams and final hackathons through Excel sheets, email threads and paper jury forms. Participant records become scattered, progress is hard to track, and each new cohort means rebuilding the same process from scratch. The biggest failure point is hackathon judging: scores are collected manually, calculated by hand, and lack a clear audit trail. When teams question the outcome, managers cannot show which jury member gave which score against which criterion. As a result, they cannot confidently prove participant development or report measurable program impact to funders and university leadership.

Open the demo Video GitHub

Preliminary score

59.5/ 100Promisingaverage of 2 judges

Value for the user19.0 / 25
A specific problem and a clear outcome from solving it.
Prototype and use of AI10.0 / 30
A working core scenario, and what AI actually contributes to the result.
Quality testing14.5 / 20
Tests, examples of failures, and comparison with the current approach.
Feasibility11.0 / 15
Data requirements, running costs and a clear next step.
Originality5.0 / 10
A meaningful difference from common solutions.

What the judges wrote

GPT-6 AstraPreliminary61/100
Strengths
Your target user and judging problem are concrete: cohort managers need traceable scores and consistent participant records, not another disconnected tool. Your testing account identifies specific concurrency, permissions and isolation failures, while honestly acknowledging incomplete coverage and unverified flows. Your itemized operating-cost assumptions and proposed real-cohort pilot provide a useful foundation for adoption planning.
To improve
Your AI contribution needs a demonstrated end-to-end scenario: the deck places AI tools on the 2027 roadmap, while the submission describes an untested mentor. Your pilot should compare judging time and reconciliation errors against spreadsheets, and evaluate the chosen AI task on representative cases with documented failures. Your production-readiness evidence should include reproducible test results, verification that tenant isolation is enforced, consistent advertised and charged prices, and costs based on the model actually used.
Claude Opus 5.5Preliminary58/100
Strengths
Your problem framing is sharp: program managers at technoparks and accelerators who cannot show which juror gave which score against which criterion is a concrete, defensible pain. Your testing write-up is unusually honest and specific. It describes a 500-concurrent-registration test for seat overbooking, a regression test against Super Admin role leakage, the move to PostgreSQL row-level security, and coverage figures that admit branch coverage falls short of your own target. Your operating-cost slide gives clear per-seat assumptions and a pilot-versus-scale breakdown.
To improve
For an AI track, AI is barely present. The AI mentor has no evaluation, the mentor, pitch coach and jury assistant are listed as Q2 2027, and your setup notes say AI tools are not complete. Build and demonstrate one AI feature end to end, such as a jury-assistant that summarises submissions against criteria, and test it on a small labelled set. Align your stated model (Claude Sonnet 4) with the cost slide (GPT-5.4 mini), and connect the USD packages to the TRY billing. Run the planned pilot cohort and measure the hours saved on judging and reporting compared with spreadsheets.

The team

Orkhan Yusifli
data
Ziyad AhmadzadaLead
product
Alsu Həsənova
business
Elvin Ahmadzada
developer
Oruc Səfərzadə
developer

Disclosure

Required by the rules. Judges verify it against the repository.

Models: Claude Sonnet 4 (Anthropic API)
Data: Synthetic demo data (seeded tenant, 100 participants, trainings, exams, simulation scenarios, sample hackathon PDFs) — no real user data yet
Components: Next.js 16, Prisma + PostgreSQL, NextAuth v5, Upstash Redis, AWS (ECS, S3, SES), Sentry, PayTR, pdf-lib
Built during the event: Confirmed

Quality testing

We wrote 62 test files covering the flows a program manager and participants go through: seat purchase, program creation, invite-link registration, seat exhaustion, payment success and failure, jury permissions, certificates, and the TR/EN/AZ interface. Line coverage on core logic is ~90%, but branch coverage is only 69.5%, below our own 80% target. What broke (found in our internal audit, now fixed and tested): - Seat overbooking: when many people registered at once, more participants than the cohort's seat limit could get in. We fixed this with row locking, and a test now fires 500 simultaneous registrations and checks that exactly the seat limit succeeds. - Role leakage: the platform owner's Super Admin account could create programs and manage or score hackathons inside a customer's tenant. Those permissions were removed, and a test fails if they come back. - No data isolation: the first build stored all organizations in one SQLite database with no separation. We moved to PostgreSQL with row-level security. Still broken or unverified: - Prices: the displayed packages are in USD, but billing plans are in TRY (PayTR), and the two are not connected. We cannot yet guarantee that the advertised price is what gets charged. The live payment flow is untested. - AI mentor: it has no automated tests or evaluation set, so the quality of its answers is unverified. - Email templates exist only in Azerbaijani, and personal-data export/deletion is not built yet. - Everything was tested on synthetic demo data, not with a real cohort. Compared with today's spreadsheets, email and paper jury forms, BizSim keeps participant records in one place, totals jury scores automatically and stores who gave which score against which criterion. We have not measured time saved or learning outcomes yet. A pilot with one real cohort is our next validation step.

Worth 20 of the 100 points.

Running it

Use these credentials: - **Super Admin** - Email: [email protected] - Password: s7zvc6tkfmY+PdTbKEZGNExjS4+Yikg0 - **Organizer / Tenant Admin** - Email: [email protected] - Password: Tenant123! The Super Admin manages tenants, billing, subscriptions, and audit logs. The Organizer account manages programs, participants, training, hackathons, jury assignments, and reports. Known limitation: this is a demo environment; AI tools, GDPR/KVKK data-export and deletion flows, and production payment verification are not yet complete.

Project ownership

The team keeps the rights to this project. Judges were given access for verification only.