Spark Program | VibeQuest: Turning AI-Assisted Coding Into Real Learning

1. Project Overview

Project Name: VibeQuest

One-Sentence Summary:
VibeQuest turns AI-assisted coding into structured learning by making users study, verify, and defend generated CKB/Fiber code before completing a quest.

Project Type: AI Learning Platform / Gamified CKB Onboarding Product / CKB-Fiber Education DApp

Live Product and Repositories:
Frontend Demo: https://vibequest-web-seven.vercel.app
Frontend Repository: https://github.com/buidlLabs3/vibequest-web
Backend Repository: https://github.com/buidlLabs3/vibequest-core
Product Architecture: https://github.com/buidlLabs3/vibequest-web/blob/main/docs/product-architecture.md
Recorded Walkthrough: https://github.com/buidlLabs3/vibequest-web/blob/main/demo.webm

2. Builder Profile

VibeQuest is being built by xbeach. I handle product direction, frontend, backend, AI learning flow, CKB/Fiber integration direction, deployment, and demo preparation.

Contact email: [email protected]
Telegram: @OXbeach
GitHub organization: https://github.com/buidlLabs3

The current product already has a deployed Next.js frontend, Rust backend, OpenAI-powered learning and quest generation, MongoDB persistence, wallet-based identity flow, a dashboard, code workbench, and ship-gate structure. The Spark request is to turn this working prototype into a sharper, reviewable MVP.

3. Product Background

AI-assisted coding is now part of how people build. Many new builders ask AI to generate code, run it, modify it, and continue. That workflow is not going away, but it creates a serious gap: people often ship code they cannot explain.

That gap is especially risky in CKB and Fiber. CKB uses cells, lock scripts, type scripts, witnesses, transaction structure, and asset logic. Fiber introduces payment channels, invoices, payment proofs, and off-chain state assumptions. A learner can ask AI to generate a paywall, payout split, receipt verifier, or cell proof flow, but still fail to understand what the code trusts, what can be replayed, and what must be verified.

VibeQuest makes AI-assisted building safer by turning it into a learning loop: generate, inspect, test, explain, and record progress.

4. Focused MVP Scope

The current VibeQuest MVP is focused on two narrowed learning areas: CKB Cell Model and Fiber Payment Models.

The CKB Cell Model path teaches how cells, OutPoints, lock/type scripts, witnesses, transaction evidence, wallet proof, and verifier logic connect to generated code. The goal is to help learners understand what a CKB-based app is actually trusting before they accept or modify AI-generated code.

The Fiber Payment Models path teaches how invoices, payment proof, channel state, receipt binding, replay risk, and paid-access logic connect to practical application flows. The goal is to help learners understand how payment-related evidence should be scoped to the right user, content, run, cell, amount, and state.

These two areas form the core MVP because they cover the most important learning problem VibeQuest is solving: helping users move from AI-generated output to real understanding of proof boundaries.

The current product loop is:

AI lesson → checkpoint → lesson-based code quest → generated verifier/test files → workbench verification → boss challenge → completion record

5. Product Solution

VibeQuest is a gamified learning workbench where users learn by doing. A learner selects a path, generates a structured lesson module, studies the concept, answers a checkpoint, asks the AI tutor questions, writes notes, unlocks a practical quest, inspects generated code, runs verification checks, solves a boss challenge, and records progress.

The product’s main idea is simple: AI is the execution layer, but understanding is the progression layer.

A user can use AI to generate code, but VibeQuest only rewards progress when the user proves they understand the generated code. The boss challenge is tied to the generated verifier and denial tests, not generic trivia.

For the first MVP path, learners should answer questions like: What does this verifier trust? Is the trusted value coming from CKB transaction evidence, wallet proof, or frontend JSON? Which field would an attacker mutate first? Does the denial test prove replay prevention?

6. User Experience and Architecture

A user opens VibeQuest and starts by choosing a learning path, speciality, goal, and pace. The current MVP is centered on CKB Cell Model and Fiber Payment Models, but the learner experience is personalized by background: vibecoder, backend developer, frontend developer, security auditor, or product/community learner.

VibeQuest is not a static quiz site or a repository of pre-written lessons. The learning structure is fixed enough to keep the product reviewable, but the actual lesson content, checkpoint questions, tutor answers, code quests, denial tests, code explainers, and boss challenges are generated dynamically by AI from the learner profile and active lesson.

The product flow is:

Learner profile → AI-generated learning path → checkpoint → lesson-based code quest → generated verifier/test files → Workbench verification → boss challenge → completion record

Each generated learning path contains five modules. Each module includes a deep explanation, code lens, checkpoint question, wrong-answer feedback, tutor context, reference trail, and a quest bridge. The learner must pass the checkpoint before generating the related practice quest.

Once unlocked, the quest is generated from that completed lesson’s concept, checkpoint, misconception, and practice bridge. In the Workbench, the learner inspects generated files, runs file checks, asks questions about the code, and answers a boss challenge tied to the generated verifier and denial tests.

The frontend is built with Next.js and TypeScript. It owns the learning interface, dashboard, workbench, wallet flow, and ship gate. The backend is built with Rust and Axum. It owns AI generation, output validation, wallet proof validation, persistence, quest state, and reward-claim preparation. MongoDB stores users, learning sessions, generated lessons, quest runs, tutor messages, notebook records, completion states, and reward claims.

Product architecture: https://github.com/buidlLabs3/vibequest-web/blob/main/docs/product-architecture.md

7. Technical Approach and Current Progress

VibeQuest is built as a working AI learning product with a Next.js frontend and Rust backend. The product currently supports AI-generated learning paths, interactive checkpoints, tutor questions, lesson-based code quests, a code Workbench, generated file verification, boss challenges, MongoDB activity records, wallet-based identity, and a recorded demo walkthrough.

The backend treats AI output as a draft artifact, not as trusted product content. A generated lesson or quest must match the required structure before it is accepted. For lessons, this means five coherent modules with clear explanations, checkpoints, misconceptions, and practice bridges. For quests, this means generated implementation files, test files, denial paths, CKB/Fiber proof signals, a code explainer, and a boss challenge tied to the generated code.

If the AI response is incomplete, too generic, or does not meet the required structure, VibeQuest does not silently load placeholder content. The user is asked to regenerate. This keeps the MVP honest and easier to evaluate: every accepted quest must be specific enough for a learner to inspect, test, and defend.

The core technical principle is that learners should not only receive AI-generated code, they should prove they understand the trust boundary behind it. A completed quest must show that the learner can explain what the verifier trusts, what can be replayed or mutated, and which denial test proves the unsafe path is blocked.

8. Funding Request and Budget

Total requested: $1,000

Product and development work: $600
Frontend refinement, backend hardening, AI learning flow improvements, quest generation improvements, dashboard organization, workbench verification, and end-to-end testing.

OpenAI/API usage: $150
Lesson generation, quest generation, tutor interaction, repeated testing, and demo preparation.

Hosting and infrastructure: $100
Vercel hosting, backend deployment, MongoDB usage, and operational testing during the grant period.

CKB/Fiber polish and QA: $100
CKB/Fiber reference integration, wallet flow polish, ship-gate clarity, and ecosystem-specific testing.

Documentation and demo material: $50
README verification guide, walkthrough preparation, architecture documentation, proposal revision, and demo recording.

The AI cost model is controlled by design. VibeQuest does not call AI on every page load or refresh. AI calls happen when a learner generates a new learning path, asks the tutor, turns a completed lesson into a quest, or creates a custom quest. Accepted lessons, quests, tutor notes, and progress records are stored in MongoDB, so users can resume and review without repeated model calls.

After the Spark grant, ongoing costs will be managed through compact prompts, persistence, rate limits, low-cost hosting tiers, and usage monitoring. If beta usage grows, the project can support ongoing operation through ecosystem sponsorship, beta cohorts, or paid learning tracks.

9. Milestones and Deliverables

Milestone 1: CKB Cell Learning Path and Dashboard

Funding requested: $300

This milestone focuses on one learning path: Understanding CKB Cells and Proof Boundaries.

Deliverables:

Five dynamically AI-generated lessons for the CKB Cell path, created from the learner’s selected profile, goal, speciality, and pace. These are not static lessons.. Each lesson includes explanation, code lens, checkpoint, correct answer, wrong answers, feedback, tutor context, and quest bridge. Dashboard shows completed lessons, incomplete lessons, tutor questions, notebook reflections, and related quests. Learners can resume a saved learning path.

Acceptance criteria:

A reviewer can generate the CKB Cell path, see at least five modules, complete a checkpoint, ask a tutor question, write a note, and confirm the dashboard records the activity.

Milestone 2: Lesson-Based Quest Workbench and Verification

Funding requested: $350

This milestone turns completed lessons into practical code quests generated from the lesson’s concept, checkpoint, misconception, and practice bridge.

Deliverables:

Lesson-to-quest generation from completed lessons. AI-authored implementation file and test file. At least one valid test case and at least two denial-path test cases. Workbench display for generated files. Generated file verification checks. Code explainer tied to active generated files. Boss challenge tied to the generated verifier and denial test.

Acceptance criteria:

A reviewer can complete a lesson, generate a related quest, open generated code in Workbench, run file checks, answer a boss challenge, and see the quest completion state update. The boss challenge must test understanding of the generated code, not generic CKB trivia.

Milestone 3: CKB/Fiber Polish, Verification Guide, and Public Demo

Funding requested: $350

This milestone makes the MVP easy for reviewers and the CKB community to test.

Deliverables:

Clear public demo flow for one complete CKB Cell learning path. Recorded walkthrough showing lesson generation, checkpoint, quest generation, workbench verification, boss challenge, and completion. README verification instructions. Product architecture documentation. Clear CKB/Fiber reference links in the product. Ship gate state that explains what is complete, pending, and reward-claim ready. Frontend and backend checks passing.

Acceptance criteria:

A reviewer can follow the public demo or recorded walkthrough and verify the complete learning loop without private instructions.

10. Verification Plan

A reviewer can verify the product through this flow:

Open the frontend demo, connect wallet, choose the CKB Cell learning path, generate the five-module learning path, open a lesson, answer the checkpoint, generate the lesson-based quest, inspect generated files in Workbench, run generated file checks, answer the boss challenge, confirm completion state, and open the dashboard to confirm progress was recorded.

For technical verification:

Frontend: npm run lint and npm run build
Backend: cargo test
Repositories: public GitHub repos
Walkthrough: https://github.com/buidlLabs3/vibequest-web/blob/main/demo.webm

Expected output: the learning path appears with five modules, the quest is generated from a completed lesson, generated code and tests appear in Workbench, verification checks run, the boss challenge is tied to the generated code, and completion is recorded in the dashboard.

11. Current State vs. Funded Work

The current state is a working prototype with deployed frontend, deployed backend, AI generation, MongoDB persistence, wallet identity flow, learning modules, quest generation, workbench checks, dashboard records, and ship gate structure.

The funded work is the product-quality layer. It narrows the MVP, improves clarity, improves learning depth, strengthens generated quests, improves verification, and makes the demo easy for reviewers to test.

The main difference after this revision is focus. Instead of claiming broad ecosystem coverage immediately, VibeQuest will prove one complete CKB learning path first, then expand.

12. CKB Ecosystem Alignment and Final Summary

VibeQuest supports CKB onboarding by helping learners understand the concepts that matter when building real applications: cells, OutPoints, lock scripts, type scripts, witness data, transaction evidence, wallet proof, replay risk, Fiber invoice context, and proof-bound reward claims.

Many future builders will enter ecosystems through AI-assisted workflows. If CKB wants more builders, it also needs learning tools that meet users where they are. VibeQuest turns AI-assisted coding into a structured path toward understanding, safer development, and better ecosystem onboarding.

The requested $1,000 supports three clear milestones: learning path and dashboard, quest workbench and verification, and CKB/Fiber polish with public demo material.

3 Likes

Hi @XBeach,

Great to see you have interested in Spark Program!

Below are some personal thoughts before I submit this for committee review, for your reference. These reflect my own reading and do not represent the committee’s position.

1. Scope & Priority:

The proposal identifies a growing problem: vibe coding without comprehension is risky, especially in CKB’s Cell model and Fiber’s off-chain state complexity.

However, the scope maybe is ambitious for a MVP. I recommend focusing on one well-defined learning path (e.g., CKB basics or Fiber payment channels) rather than attempting full ecosystem coverage in the first milestone.

2. Deliverable Clarity:

The proposal lists milestones (e.g., learning content generation, quest/challenge system, workbench, ship gate), but the deliverables for each milestone could be more concrete. For example:

  • Milestone 1: “AI-generated learning content” — what is the expected output? A set of 5–10 lessons? A structured curriculum outline? A working lesson page with AI-generated text and code snippets?
  • Milestone 2: “Quest/challenge system” — what does a “Boss challenge” look like? Is it a multiple-choice quiz, a code-editing task, or a live transaction simulation?

Please clarify the exact artifacts (e.g., number of lessons, types of challenges, test cases) and how they will be verified by the committee.

3. Verification Method:

The product already has a deployed demo (frontend + backend), which is excellent. However, the current demo is primarily a landing page with limited interactive learning flow. To verify the core value proposition, the committee would need to:

  • Experience a complete learning loop: from AI-generated lesson → quest → code verification → boss challenge → completion.
  • Validate the accuracy of AI-generated CKB/Fiber content: e.g., does the AI correctly explain Cell model concepts, transaction structure, or HTLC proof flows?
  • Test the “proof-of-understanding” mechanism: does the boss challenge genuinely test understanding, or is it a superficial quiz?

I suggest you deploy a fully functional end-to-end prototype for at least one learning path (e.g., “Understanding CKB Cells”) and share a public link or a recorded walkthrough.

4. Budget Justification:

The currently requested $1,000 budget does not exceed the $1,000 cap for pure technical projects. However, the budget breakdown is not provided. Please clarify:

  • What are the main cost drivers (e.g., OpenAI API usage, hosting, domain, developer time)?
  • Is there any plan to cover ongoing operational costs (e.g., API fees, database hosting) after the Spark grant ends?
    The committee may consider adjusting the budget if the scope is narrowed.

Please revise and reply in this thread. I’m happy to help clarify any questions or provide additional feedback.

Best,
xingtian

4 Likes

Hi @xingtianchunyan,

Thank you for the thoughtful feedback. I revised the proposal with your points in mind.

The scope is now narrowed around a focused MVP instead of broad ecosystem coverage. The current MVP centers on two practical learning areas: CKB Cell Model and Fiber Payment Models. The goal is to prove one complete learning loop clearly: AI lesson → checkpoint → lesson-based code quest → generated verifier/test files → workbench verification → boss challenge → completion record.

I also made the deliverables more concrete. The revised draft now clarifies the expected lesson modules, checkpoints, code quest artifacts, denial-oriented tests, Workbench flow, boss challenge behavior, completion records, and how each milestone can be reviewed.

The Learning tab is now positioned as the main product flow. It lets a learner generate a focused lesson path, study CKB/Fiber concepts, answer checkpoints, ask tutor questions, and then turn completed lessons into practical code quests. The Workbench then verifies generated code, tests, and understanding.

I also added a clearer budget breakdown for the $1,000 request, covering development time, OpenAI usage, hosting/infrastructure, QA/polish, and demo/documentation work, plus a note on keeping the MVP lightweight after the Spark grant.

Links:
Frontend repo: https://github.com/buidlLabs3/vibequest-web
Backend repo: https://github.com/buidlLabs3/vibequest-core
Demo walkthrough: https://github.com/buidlLabs3/vibequest-web/blob/main/demo.webm

I appreciate the guidance. The revised draft should now be easier for the committee to evaluate against scope, deliverables, verification, and budget.

3 Likes

gm gm @xingtianchunyan hope you had time to review the updated proposal and response

2 Likes

Hi @XBeach ,

Thank you for your enthusiastic participation in the Spark Program! I have reviewed the updated proposal and responses, and have set the proposal status to “Submitted.”

As soon as the committee issues its official review opinion, I will respond to you here as soon as possible.

Best,
xingtian

2 Likes

Thank you for the response, looking forward to hearing from you soon.

1 Like

Hi @XBeach ,

After review by the Spark Program committee, your proposal has been assigned a status of Pending. This is not a rejection, but an invitation to revise.

The committee recognizes the potential of your project as an educational tool for developers learning CKB and Fiber.

Main Revision Recommendations:

  1. Clarify the underlying technical architecture:
    Please specify whether the quest content (questions, challenges) is pre-defined static content or dynamically generated by AI in real-time. Your proposal mentions “OpenAI-powered learning and quest generation”, but a clear architecture description will help the committee assess feasibility and scope.

  2. Maintenance Cost & Personalization (if AI real-time generation):

  • If the questions are AI-generated in real-time, can the ongoing maintenance and API costs be sustainably covered within the grant and post-grant operational plan?
  • Can different users customize settings based on their own preferences, skill levels, or learning goals? If so, please describe the personalization mechanism.

Please revise and resubmit based on the feedback above. We will arrange a new round of review as soon as possible.

Looking forward to your reply.

Best,
xingtian
On behalf of the Spark Program Committee

1 Like

Hi @xingtianchunyan and Spark Program Committee,

Thank you for the clear review. I have updated the GitHub product architecture note to make the AI flow, personalization model, and cost controls easier to evaluate:

https://github.com/buidlLabs3/vibequest-web/blob/main/docs/product-architecture.md

To clarify the architecture: VibeQuest is not a static quiz site or a repository of pre-written lessons. The product uses constrained learning lanes, but the actual lesson content, checkpoint questions, tutor answers, code quests, denial tests, code explainers, and boss challenges are generated dynamically from the learner’s selected profile and active lesson.

The flow is:

Learner profile → AI-generated learning path → checkpoint → lesson-based code quest → generated verifier/test files → Workbench checks → boss challenge → completion record.

The backend does not pass arbitrary AI output directly into the product. vibequest-core validates generated lessons and quests against a required structure before accepting them. A valid quest must include generated implementation files, denial-oriented tests, CKB/Fiber proof signals, a code explainer, and a boss challenge tied to the generated code. If the output is incomplete or generic, the user is asked to regenerate.

On maintenance cost: real-time AI is used only at learning or quest creation moments, not on every page load. Generated lessons, quests, tutor notes, progress, and completion records are stored in MongoDB, so users can refresh, resume, and review without repeatedly calling the model. This keeps the MVP cost controlled during the grant period and makes post-grant operation realistic through low-cost hosting, usage limits, ecosystem sponsorship, beta cohorts, or paid learning tracks if adoption grows.

On personalization: users choose a learning lane, speciality, goal, and pace. A vibecoder gets more help understanding what generated code trusts. A backend developer gets deeper verifier logic and denial tests. An auditor gets replay, mismatch, and failure-path challenges. A product/community learner gets clearer conceptual framing without forcing them into heavy implementation details. Checkpoint results and tutor questions also become learning evidence that can shape later quests.

The key evaluation point is one complete learning loop: can a CKB/Fiber learner move from AI lesson, to understanding check, to generated code, to verification, to proof-of-understanding. That is the MVP I am pushing VibeQuest toward.

1 Like

Hello just posting this here, I was readjusting the proposal based on the above response and got edit conflict while at it. I’ve saved the changes though, I hope I’ll be able to publish the small tweaking I was making, thank you.