AI Enterprise SolutionsGPT & Claude
TinyTamers
We are TinyTamers and our project AVEN aimed at helping people with disabilities and service centers communicate more easily. We address communication barriers, limited accessibility, and the difficulties people with disabilities face when accessing everyday services independently.
Preliminary score
76.5/ 100Strongaverage of 2 judges
- Value for the user18.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI24.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing15.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility12.0 / 15
Data requirements, running costs and a clear next step.
- Originality7.5 / 10
A meaningful difference from common solutions.
What the judges wrote
- GPT-6 AstraPreliminary75/100
- Strengths
- Your documented prototype addresses a concrete service-desk communication barrier with a custom 60-class sign recognizer and a 100-word sign-video return path. Your validation report distinguishes 95.9% accuracy on answered clips from 40% coverage and openly acknowledges shared training/validation signers and confusing sign pairs. Your landmark-only camera pipeline, uncertainty handling and explicit limits on medical/legal use are thoughtful safeguards.
- To improve
- Your next evaluation should include held-out signers and Deaf users, comparing task completion, misunderstandings and service time against current communication methods. Demonstrate a complete two-way service-desk exchange, including recovery from abstentions and unsupported words. Reconcile the claimed 92% scenario success with the reported validation metrics by specifying sample sizes and what each measure represents. Clarify server-versus-browser inference and estimate hosting, LLM and speech costs beyond free tiers.
- Claude Opus 5.5Preliminary78/100
- Strengths
- Your core scenario is real AI work. In-browser MediaPipe landmarks feed a GRU classifier you trained yourselves on AzSLD, and the result is an LLM sentence plus Azerbaijani TTS, with a reverse text-to-sign video path. The 'not sure' abstention rule is a smart design choice: answering 40% of clips at 95.9% accuracy is a sensible trade-off for a service desk. Your README is unusually honest about its limits, including validation leakage because no signer IDs exist, confusion between MƏN / MƏNİM / MƏNƏ, and isolated signs only. The privacy design of sending landmarks only and keeping camera images in the browser is concrete and well explained.
- To improve
- The '92% success rate in test scenarios' in your submission has no stated test set, and it does not match your 69.5% top-1 / 95.9%-when-answering figures. The automated test count also varies between 245 and 266, so define your test scenarios and report one consistent set of numbers. Run your new-signer protocol on team recordings and on at least a few deaf signers, because that is the number that decides real-world usefulness. For the enterprise track, compare against today's workflow: interpreter availability, wait time, and the share of service-desk interactions completed without an interpreter. Your user persona ('Disable People') is too broad, so focus on one desk scenario such as ASAN xidmət with a concrete script of the signs it needs.
The team
Gasimiva AyselLead
designer
Elif Feyziyeva
developer
Rashid Dursunov
developer
Ulviyya Aliyeva
designer