Neo
Our platform is designed for junior developers, startup teams, enterprise architects, and engineering leaders who want to design better software systems and collaborate with AI agents.AI has made software development faster, but designing reliable and scalable systems remains a challenge. Junior developers and startups often lack architectural experience, while even large companies can overlook critical issues during code and architecture reviews. These mistakes can lead to system failures, security risks, and expensive maintenance.
Preliminary score
- Value for the user16.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI21.5 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing9.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility12.0 / 15
Data requirements, running costs and a clear next step.
- Originality7.0 / 10
A meaningful difference from common solutions.
What the judges wrote
- Claude Opus 5.5Preliminary61/100
- Strengths
- Your README describes a substantial working system rather than a mock-up. It includes a React Flow architecture canvas, tool-calling agents that ground answers in the canonical graph, proposals that apply nothing until an owner or admin approves, and audit trails. The ChatGPT + Gemini council, with sandboxed rooms, shared token and time limits and the 4-of-5 decision rule, is a thoughtful design for keeping AI review bounded and human-owned. You also measured real running costs in LangSmith (83 traces, $2.55, about $0.03 per trace, P50 3.05 s), which is far better than estimating them. Your seeded findings are explicitly tied to declared configuration rather than invented telemetry.
- To improve
- Your testing so far covers infrastructure (the 1,000-request load test and the crash that led to adding a load balancer), but not whether the agents give good architecture advice. Seed a few architectures with known flaws, measure how many each agent catches and how many false findings it raises, and compare this with a human review or a single-chat assistant. Narrow your user from 'juniors to enterprise architects' to one persona, and name the metric a company would adopt this for, such as review time saved or defects caught before production. Align your naming (Neo / AI Space / AgentSpace) and your deck's 'prototype in progress' status with what the README says actually works.
- GPT-6 AstraPreliminary70/100
- Strengths
- Your design connects specialist architecture reviews to concrete components, evidence-backed findings, and human-approved changes rather than leaving recommendations in chat. Your README provides a reproducible seeded setup and clearly distinguishes deterministic local agents from model-backed, tool-calling agents. Reporting the 1,000-request crash, remaining rate-limiting gap, and measured LangSmith usage makes important operational limitations visible.
- To improve
- Your deck still describes an unfinished prototype, so document one completed model-backed path from system description to reviewed architecture, approved change, and saved decision record. Choose one initial customer segment and compare review time and risk-detection accuracy against its current process, including false positives and missed risks. Report cost and latency per complete review, since per-trace averages do not establish the expense of multi-agent sessions. Repeat a documented concurrent load test after adding rate limits and verify whether the load-balancing fix prevents the crash.
The team
Mahammad Rza Cahangirov
developer
Ziya Gasimov
developer
Ulvi SharifzadeLead
data