AI Enterprise SolutionsGPT & Claude

ATA Team

Automated SLA compliance monitoring & AI-powered refund recovery for SaaS/Cloud vendors.

Open the demo Video GitHub

Preliminary score

58.5/ 100Promisingaverage of 2 judges

Value for the user16.5 / 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 testing12.5 / 20
Tests, examples of failures, and comparison with the current approach.
Feasibility8.5 / 15
Data requirements, running costs and a clear next step.
Originality6.0 / 10
A meaningful difference from common solutions.

What the judges wrote

Claude Opus 5.5Preliminary57/100
Strengths
Your SLA engine is grounded in real research: you document Cloudflare's actual credit formula, its 5-business-day claim window and its evidence requirements, and you model the claim deadline per incident. The core loop works in the demo: simulate an outage or slowdown, an incident is created, a dollar credit is estimated, and two differently-toned claim emails are logged. Your 33 node:test cases cover meaningful boundaries, such as exactly 200 ms, fewer than 5 samples, a full-month cap and zero-duration outages. Your 'Honest limits' section clearly admits the simulated metrics, the custom latency rule and the placeholder vendor address.
To improve
AI is currently optional: detection and credit math are deterministic, and emails fall back to static templates. Several claims in the README table and the deck are not shown in the build, including AI anomaly detection, root-cause analysis and the 'AI Contract Reader' that extracts SLA rules from PDFs. Either build PDF-to-SLA-rule extraction and test it on several real vendor contracts, or tone the claims down. Connect at least one real data source, such as synthetic probes against a real endpoint or the Cloudflare analytics API, and evaluate the AI claim drafts against templates for accuracy and completeness. Back the '10–25% lost credits' figure with a source or with interviews with a few finance or IT teams.
GPT-6 AstraPreliminary60/100
Strengths
Your focus on missed SLA service credits gives IT and finance teams a concrete, financially measurable outcome. Your screenshots and documented demo connect simulated outages to incident records, credit estimates, deadlines and logged notifications. Your specific boundary, authentication and fallback tests, together with explicit disclosure of synthetic metrics and the demo-only latency rule, provide useful evidence of engineering care.
To improve
Your materials should distinguish implemented AI email drafting from the unproven contract extraction, anomaly detection and root-cause capabilities advertised elsewhere. Show one real or recorded Cloudflare incident with plan-specific eligibility, required evidence and the actual generated claim text. Extend your tests to cover business-day/holiday deadlines, duplicate incidents and the $0.14 example under the applicable SLA formula, then benchmark claim preparation against manual work. Validate recovery value through a small customer pilot and report realistic per-customer operating costs beyond free-tier AI.

The team

Rufan Niyazlı
developer
Togrul Teymurov
developer
Adil AbdulkarimovLead
developer
Aykhan Fatullayev
developer

Disclosure

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

Models: OpenRouter free tier (optional) — e.g. meta-llama/llama-3.3-70b-instruct:free, mistralai/mistral-7b-instruct:free (email drafting only, falls back to static templates)
Data: Public Cloudflare SLA docs (2026-10-09); all metrics/incidents simulated (synthetic)
Components: Node/Express, React/Vite, MariaDB, mysql2, bcryptjs, jsonwebtoken, nodemailer, recharts, axios — no LangChain/Supabase/ML libs
Built during the event: Confirmed

Quality testing

What we tested SLA credit engine (sla/rules.js) Outage formula: 3-minute outage on $200 fee → exact credit $0.14 Full-month outage → proportional credit (no silent overflow) Zero-duration outage → $0 Latency rule: average of last 5 samples Exactly 200 ms → no violation 201 ms → violation + 15% of monthly fee Fewer than 5 samples or null/empty → safe no-violation AI prompt construction (ai.js) Company and vendor prompts always contain the real incident facts (type, duration, credit amount, claim deadline) Vendor prompt explicitly references Cloudflare’s 100% uptime SLA and 5-business-day claim window No unresolved placeholders ({{...}} or trailing [...]) Company prompt stays under the 120-word instruction budget AI resilience generateText() returns null when OPENROUTER_API_KEY is missing → system falls back to static templates without crashing the pipeline Auth & schema Basic auth middleware and database schema integrity checks All tests run with the standard Node test runner: Bashnode --test backend/tests/*.test.js What broke / what we deliberately exposed Real Cloudflare API is not connected yet — metrics are simulated. The monitoring loop and incident creation work, but they run on synthetic data. AI is optional — without an OpenRouter key the system continues with static templates. This is intentional graceful degradation, not a crash. Latency rule is a custom demo rule — real Cloudflare SLA has no latency clause. The UI labels it as such so judges are not misled. No end-to-end integration test that actually sends mail or hits a live vendor endpoint (we keep MAIL_MODE=log for safety). Claim emails are generated and logged, but no real money is moved and the vendor address is still a placeholder. These limitations are documented in the README under “Honest limits (MVP)” and are visible in the 3-minute demo flow.

Worth 20 of the 100 points.

Running it

https://github.com/photomanai/Recoup/blob/main/SETUP.md

Project ownership

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