DDS Building Agentic AI Application Challenge 2026 · Day 5

Understand the math, not just the answer.

Nevin Chatbot helps students learn mathematics through concept explanations, structured step-by-step guidance, verification and interactive follow-up questions.

Start learning Open live Gemini app ↗

Mathematics only Gemini-powered Free for students & teachers No credit card required No API key required from users

Day 5 · Integrated user interface

Ask a mathematics question

The Day 5 interface connects user input, validation, math-only scope control, the secure Gemini API path, formatted output, progress feedback, and a direct fallback link to the working Google AI Studio deployment.

Learning level

Math-only scope Non-mathematics questions are politely refused.

Your math question

0/3000 characters · approximately 0 input tokens. Do not enter personal or sensitive information.

AI-generated mathematical explanations can make mistakes. Verify important calculations and academic work.

Tutor conversation

Ready when you are

Math tutor ready

Latency—

Format check—

PrivacyNo PII needed

ƒ(x)

Ask a math question to begin. Nevin Chatbot will explain the concept, show the steps, and help you verify the result.

Seven-step LLM project workflow

From a real problem to a reliable learning product

01Problem
Students often receive answers without understanding.
02Use case
Teach math with clear steps, verification and follow-ups.
03Prompt
Define scope, structure, learning level and refusal behavior.
04Tools
Use deterministic calculation and equation verification.
05Prototype
Connect the interface to the secure model endpoint.
06Evaluate
Measure accuracy, coherence, refusal quality and latency.
07Deploy
Publish, monitor, collect feedback and improve.

Day 5 UI testing

Test the whole system, not only the model

The included evaluation harness covers mathematics accuracy, response structure, safe refusal of non-math questions, edge cases, and API response time. BLEU-2 is included only as a supplemental wording-similarity measure because correct tutoring responses can be phrased in many valid ways.

Accuracy

Expected-answer patterns and deterministic checks for supported calculations.

Coherence

Concept, steps, verification, final answer and learning tip structure.

Scope control

Non-math requests are refused consistently without generating unrelated content.

Latency

Response time is displayed and recorded to support real-time optimization.

Edge cases

Empty input, long input, PII, ambiguous prompts, timeouts and unavailable APIs.

Feedback

Helpful, not helpful and report-issue feedback is stored locally by default.

Security & privacy

Protect learners and provider credentials

The public frontend contains no secret key. The server-side integration validates scope and input, limits conversation history, protects API credentials through environment variables, and returns safe user-facing errors.

No API keys in GitHub Pages files
Origin restriction and request-size limits
PII and secret-pattern detection before submission
Safe text rendering with no model-generated HTML
Retries, timeout handling and local math verification
No payment form, advertising tracker or account requirement

About the project

Built through the DDS challenge

Nevin Chatbot evolved from a Day 2 interface prototype into a Gemini-powered mathematics tutor focused on responsible prompting, tool-assisted verification, evaluation, security and accessibility.

Developer: Nevin Pinto · Electrical Engineering student · AI learner and builder.

Day 5 integration decisions

Prompt engineeringUsed instead of model fine-tuning at this stage
Math-only guardReject unrelated topics
Tool augmentationVerify supported calculations
EvaluationAccuracy, format, scope and latency
Day 5 completeInterface integration, input/output handling, UI testing and export controls
Day 6 nextFinal enhancements, security and debugging