Ezam AI
AtlasFlow AI serves two primary user groups: businesses managing employee travel and individual travelers planning trips on a budget. **For businesses**, organizing a single business trip often requires switching between flight booking platforms, accommodation websites, company travel policies, approval processes, maps, and expense spreadsheets. Employees spend time comparing options manually, while managers struggle to enforce travel policies, control budgets, and approve requests efficiently. Finance teams also face difficulties separating business expenses from personal spending. **For individual travelers**, planning a trip involves researching flights, hotels or apartments, attractions, transportation, and activity costs across multiple platforms. Finding the best combination within a fixed budget is time-consuming, especially when balancing affordability, comfort, personal interests, and travel schedules. **AtlasFlow AI solves both problems through one intelligent workflow.** Users enter their destination, dates, budget, and preferences. The platform generates three optimized travel packages—Economy, Balanced, and Comfort—along with personalized itineraries, estimated cost breakdowns, interactive maps, and booking links. For businesses, AtlasFlow additionally automates corporate travel-policy checks, approval workflows, and expense reporting. Our goal is to transform fragmented, manual travel planning into a faster, more transparent, and budget-aware experience for both companies and individuals.
Preliminary score
- Value for the user17.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI15.0 / 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.
- Feasibility8.0 / 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 AstraPreliminary64/100
- Strengths
- Your corporate request-to-approval-to-CSV workflow addresses a concrete coordination problem for employees, managers and finance teams. Separating company and personal spending, while keeping financial rules deterministic, is a sensible design choice. Your testing stands out for reporting actual failures, incomplete coverage and an 11-scenario comparison against a cheapest-first baseline rather than claiming universal success.
- To improve
- Your first priority should be isolating personal/business state and revalidating existing plans after policy changes, including the documented 95-AZN hotel exceeding a revised 90-AZN cap. Add regression tests for these defects and the hydration error, then rerun failed and partial browser checks. Demonstrate verified runtime model inference and compare extraction accuracy against the deterministic fallback to establish AI's contribution. Define a small corporate pilot measuring coordination time and reporting errors, alongside supplier-data requirements and per-trip operating costs.
- Claude Opus 5.5Preliminary59/100
- Strengths
- Your testing section is unusually honest and specific: 16 browser checks with 2 failures and 2 partial passes, named defects (workspace state leakage, the 90 vs 95 AZN hotel cap not re-invalidating a plan, React hydration error #418), and a baseline comparison where your constraint planner found feasible plans in 9 of 11 scenarios versus 7 for cheapest-first. Your architecture deliberately keeps policy and price decisions deterministic and leaves language to the model, which is a sound enterprise design choice. The deployed demo covers the full corporate path from request to policy check, three packages, simulated approval, audit log and finance CSV. Separating corporate and personal budgets for one trip is a thoughtful touch.
- To improve
- AI is currently optional, and by your own admission live LLM inference has not been verified, so what judges can confirm is a deterministic planner with fallback parsing. Prove the GPT-4.1 mini intent extraction works, for example on Azerbaijani and Russian requests, and measure its accuracy against the fallback. The track asks for an adoption metric, so time one real trip request done manually versus with Ezam, or count avoided policy violations. Also add running costs per request, a plan for real inventory (e.g. a GDS or aggregator API), and a consistent product name (AtlasFlow vs Ezam).
The team
Ismayil Qafarli
business
Nicat AbbasovLead
developer
Cavad Aliyev
product