UltInstructor — AI pre-construction decision platform
User: small and mid-size residential developers in Baku and Absheron (typically 5–15 floor projects) and the investment managers who decide whether to buy a plot. Problem today: a company buys the land and locks the budget without knowing whether the idea will work. The first answer takes weeks, because the architect, geologist and estimator work separately, and alternatives are almost never compared. Baku-specific risks usually surface only after money is spent: landslide zones (Badamdar, Bayil), seismic intensity 9, shallow saline groundwater, liquefaction in sandy soil. New-build prices in Baku rose 12.4% in one year (Q2 2026, median ~2,800 AZN/m²), so every mistake costs more. Outcome with UltInstructor: the developer picks an empty plot on a satellite map of Baku, and district, size and slope fill in automatically. In under a minute they get a verdict (feasible / conditional / not feasible) with reasons, the safe floor limit and foundation type, cost, margin, ROI and the most profitable use of the plot. They also get a plan for every floor, 12 engineering systems sized to Baku conditions, a ~120-line material estimate (Excel export), a Gantt schedule with monthly cash flow, a risk register with mitigation steps, and the list of geological tests to order before buying the land.
Preliminary score
- Value for the user21.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI23.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing16.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility12.5 / 15
Data requirements, running costs and a clear next step.
- Originality7.5 / 10
A meaningful difference from common solutions.
A finalist: the live jury scored it 76.0 / 100 on stage.
What the judges wrote
- Claude Opus 5.5Preliminary81/100
- Strengths
- Your user is sharply defined: small and mid-size Baku developers deciding whether to buy a plot. Their pain is grounded in local risks such as the Badamdar and Bayil landslide zones, seismic intensity 9 and shallow saline groundwater. The prototype covers a lot of real ground: map-based plot selection with auto-filled data, a verdict with floor limit, per-floor plans, 12 engineering systems, a 125-item estimate, a Gantt schedule and a risk register, all shown on the Badamdar demo. Splitting the deterministic engine (which produces every number) from the LLM (which explains and edits parameters) is a sound design, and your AI evals check that the model never calls a failing project feasible. Your testing is honest and specific: 26 engine tests, a 900-project fuzz run, 10 real-model evals, and named bugs you found and fixed (mall revenue factor, 22 W/m² cooling load, 49 kg/m³ rebar).
- To improve
- Your engine is checked against itself and against code ranges, not against real outcomes. Run two or three past Baku projects with known actual costs, schedules and geotechnical findings through it, and report the error. The AI layer is useful but not essential to the verdict: show what the AI architect adds beyond the rule-based fallback, for example a side-by-side on the same project, or judgement on cases the rules miss. District data and prices are illustrative and static, so name the specific sources you would use for official microzoning maps and live prices, and how you would get access. A quick check with even one developer or estimator would help back up the 'weeks to under a minute' claim and show whether they would trust a concept-level verdict.
- GPT-6 AstraPreliminary80/100
- Strengths
- Your focus on Baku developers' pre-purchase decisions is specific, and the Badamdar example connects site risk, floor limits and budget to a concrete screening decision. Your separation of deterministic calculations from AI explanations and parameter edits is appropriate for this high-stakes workflow. Your testing stands out for documented mall-ranking, cooling-load and rebar errors, alongside scenario, fuzz and AI checks.
- To improve
- Your next pilot should benchmark verdicts, estimates and decision time against independent professional reports for real plots; synthetic cases and agreement between internal estimates do not establish engineering accuracy. Add traceable rule sources and uncertainty ranges to floor limits and UltScore before purchase or lending use. Demonstrate an AI chat run with before-and-after parameters and recalculated outputs, showing its benefit over the non-AI interface. Your operating-cost plan should also cover obtaining and maintaining authoritative hazard maps and current prices.
The team
Saleh Mamedov
product
Ramal İlyasovLead
developer