AI Enterprise SolutionsGPT & Claude

prodvisor.

Prodvisor is built for small business owners, entrepreneurs, and service providers who lack dedicated marketing and sales expertise. Many struggle to define their target audience, communicate their product's value, understand customer objections, and identify why conversations fail to turn into sales. Customer feedback is often scattered across messages, leaving business owners to make decisions based on assumptions rather than evidence. Prodvisor addresses this by turning business information and customer conversations into personalized, actionable growth recommendations.

Open the demo Video GitHub

Preliminary score

69.0/ 100Promisingaverage of 2 judges

Value for the user18.0 / 25
A specific problem and a clear outcome from solving it.
Prototype and use of AI22.0 / 30
A working core scenario, and what AI actually contributes to the result.
Quality testing14.5 / 20
Tests, examples of failures, and comparison with the current approach.
Feasibility9.0 / 15
Data requirements, running costs and a clear next step.
Originality5.5 / 10
A meaningful difference from common solutions.

What the judges wrote

Claude Opus 5.5Preliminary68/100
Strengths
Your evaluation is unusually honest and specific: 15 manually labelled Azerbaijani dialogues, per-dimension accuracy, objection precision and recall (100%/50%), an evidence-matching count (40/46), latency, and a prompt-injection case, all shown on an in-app test screen. Your prototype covers a coherent flow, from Business DNA onboarding to a context-aware advisor, inbox analysis that cites the source message, and conversion of findings into an action plan. The synthetic example of a customer asking about delivery and being answered on price makes the pain concrete. You also mark AI-suggested profile data as unverified, which is a thoughtful design touch.
To improve
Your outcome is still asserted: run even a small timed comparison against manual review, or a pilot with one real business, to show minutes saved or replies improved. Fifteen synthetic dialogues is a thin base. Expand to real, consented chats and focus on objection recall, which you correctly name as your main weakness. Add estimated Gemini cost per analysed conversation so the 19/49 AZN tiers can be checked against running costs. Your campaign planner and generic advisor dilute the core; tightening the story around the conversation-analysis workflow would make it more distinctive.
GPT-6 AstraPreliminary70/100
Strengths
Your prototype addresses a concrete small-business problem by connecting customer conversations, persistent business context, and follow-up actions. Your interface distinguishes unverified profile suggestions and leaves missing delivery details for the seller to clarify rather than inventing them. Your 15-case evaluation reports meaningful weaknesses—including 50% objection recall and incomplete evidence matching—and clearly separates functional results from unmeasured sales impact.
To improve
Your next validation should use consented real business conversations and measure review time and missed issues against manual review and a general-purpose AI baseline. Expand your labeled tests around missed objections, ambiguous sales stages, and evidence mismatches, publishing expected versus actual outputs for representative failures. Your feasibility plan should quantify API and hosting costs at the proposed subscription tiers and specify consent, retention, and account-isolation safeguards before moving beyond localStorage.

The team

Ilkin Alizada
developer
Nurkhan Guliyev
designer
Ilaza Javadova
designer
Matin GafarovLead
product

Disclosure

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

Models: Gemini 3.5 Flash Lite
Data: Team-created synthetic customer conversations and fictional business profiles. The evaluation dataset contains 15 manually labeled Azerbaijani conversations covering customer intent, objections, sales stages, unanswered questions, and prompt injection. Users can also provide their own business information and import conversations. No proprietary third-party dataset or private customer database is bundled with the application.
Components: Next.js 15, React 19, TypeScript, Tailwind CSS, Lucide icons, Zod validation, and the Google Gemini API. Custom Business DNA onboarding, customer intelligence, business advisory, campaign planning, and evaluation modules. Browser localStorage is used for MVP data persistence.
Built during the event: Confirmed

Quality testing

We tested Prodvisor using 15 manually labeled, synthetic Azerbaijani customer conversations covering pricing objections, trust, delivery, purchase intent, unanswered questions, ambiguous conversations, and prompt injection. Measured results: - 15/15 test cases executed successfully. - Customer intent classification: 93%. - Sales stage classification: 73%. - Exact objection detection: 60%. - Objection precision: 100%; recall: 50%. - Supporting evidence matched to source text: 40/46. - Average analysis time: 2.7 seconds. - Prompt injection test: 1/1 passed. Failures identified: The model missed some valid customer objections, misclassified certain sales stages, and did not consistently reproduce all supporting evidence. Objection recall was the clearest weakness. Comparison with the current approach: Business owners typically review conversations manually or use general-purpose AI tools without a persistent business profile. Prodvisor combines structured analysis, business-specific context, supporting evidence, and recommended actions in one workflow. A controlled time-saving comparison against manual analysis has not yet been completed. Next steps: Improve objection recall, strengthen evidence validation, expand the evaluation dataset, and test with consented real-world business conversations.

Worth 20 of the 100 points.

Running it

1. Clone the GitHub repository. 2. Run npm ci. 3. Create .env.local using .env.example. 4. Set GEMINI_API_KEY and GEMINI_MODEL=gemini-3.5-flash-lite. 5. Run npm run dev and open http://localhost:3000. For the demo, complete Business DNA onboarding or use the synthetic sample business. Import sample conversations through the Inbox and run customer analysis. Known limitations: Instagram/Messenger live synchronization and payment processing are not implemented. Business profiles and demo conversations are stored in browser localStorage.

Project ownership

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