Learn anything. Your way.
Pantha is an AI-native learning platform where learners describe what they want to study, answer a few clarifying questions, and receive a structured course from beginner to advanced. Along the way they earn non-transferable experience points, compete for social streaks with friends, chat with end-to-end encryption, buy perks with a tradable token, and—on chain—can anchor verifiable course-completion credentials.
Many platforms offer fixed curricula or generic “AI courses.” Pantha focuses on personalization: the model infers your goal, resolves uncertainty with follow-up questions, and either matches you to an existing course in the vector index or generates a new course with a full chapter outline. Each chapter is turned into interactive pages (quizzes, true/false, matching, fill-in-the-blanks, image-based tasks, teach-and-explain content, and more). Your answers are scored, explanations can be generated for the last attempt, and progress advances through the course.
A social layer sits on top: follow others, become friends when mutual follows exist, see profiles and enrollments (with privacy controls), and maintain personal learning streaks and friend streaks that depend on both people being active on the same calendar day. Real-time updates (new DMs, streak milestones) reach the client over Server-Sent Events.
The economy separates reputation from money: $PXP is soulbound points minted when you learn; $PANTHA is an ERC-20 with permit support used for purchases and for reward distribution to learners proportional to their PXP-weighted share—similar in spirit to MasterChef-style staking rewards, but with XP as the weight instead of staked LP tokens.
- Guided intake: category, free-text intent, and iterative clarification questions with a configurable budget so the course matches skill and interests.
- Semantic match or create: embeddings and a vector database find similar existing courses; otherwise the system authors a new course skeleton (title, description, topics, ordered chapters).
- Rich chapters: async generation of chapter pages; AI-generated illustrations where appropriate; correct answers are not leaked to the client.
- Adaptive feedback: optional LLM explanations after answers, tuned for a mobile-friendly learning flow.
- Progress: enrollment, chapter ordering, and completion/revision tracked so first-time completion can mint more XP than practice runs.
- Non-transferable ERC-20: minted through the orchestrator, transfers and approvals disabled so XP stays tied to the wallet that earned it.
- Earned by learning (e.g. completing or revising chapters); amounts scale with how well you did on scored activities.
- Transparent reasons on-chain (short-coded “reason” plus a resource id tied to the chapter), so indexing and analytics can attribute mints.
- Tradable token with EIP-2612 permit for gas-conscious checkouts in the shop.
- Reward vault logic: the protocol tracks total XP and each user’s XP balance as shares. When the platform (or treasury workflow) deposits PANTHA into the distributor, everyone’s share of the next chunk of rewards is proportional to their XP relative to the network—the same mathematical idea as MasterChef
accRewardPerShare, applied here to minted XP instead of farm deposits. Users can claim accrued rewards; minting new XP also updates their bookkeeping so rewards stay fair.
- In-game purchases priced in $PANTHA: e.g. Streak Freeze (shield a missed day conceptually; policy prevents abuse like duplicate unconsumed purchases where enforced).
- Gas-light UX: purchases use permit so users sign once instead of separate
approvetransactions when supported. - Revenue routing: payments go to the treasury contract; the product can fund weekly or event-driven reward rounds from that flow.
- Roadmap-friendly: an XP multiplier or similar boost fits the same shop +
bytes8item id pattern as streak freeze.
- Personal streak: extends when you’re active on consecutive local days (respecting your timezone); multiple activities the same day do not inflate the count; missing a day resets the pattern.
- Friend streak: only grows when both friends record activity on the same day; one-sided activity pauses the mutual streak until both show up again.
- Wallet-derived keys and ECDH to agree on shared secrets; symmetric encryption for message bodies.
- On-chain key directory: users can register salt + public key material via a signed EIP-712 message so others can encrypt for them without a centralized key server.
- Direct messages stored as ciphertext; recipients decrypt client-side. Message policy per user: who can DM (anyone, friends only, or no one).
- Live notifications when a new DM arrives.
- Follow / unfollow, followers, following, and mutual friends.
- User search, profiles (name, username, visibility), and gated views for private profiles.
- Feed & leaderboard (product vision) — The app includes Social experiences such as a community feed and friends leaderboard in the UI; full backend feed posts and a dedicated ranked leaderboard API are still evolving—XP totals and on-chain pending rewards already give a fairness signal for “who earned most” style competition.
- Certification authority on chain commits a Merkle root of a learner’s action chain (hashed log of learning actions), enforces one-time use of each root, and authorizes minting of ERC-721 certificates with metadata URIs.
- This gives a path to tamper-evident, verifiable completion without exposing the entire private learning trail on chain—only the committed root and the NFT proof.
- Sign-In with Ethereum-style flow: nonce challenge, signed message, JWT session for the API.
- Test faucet for $PANTHA (cooldown-gated) so demos and hackathon judges can try purchases and flows without mainnet funds.
- Background jobs for long AI or content-prep steps, with job status polling for a responsive UX.
Application
- Bun as the JavaScript runtime for the API and tests
- Hono HTTP framework, Zod validation
- React 19, Vite, Tailwind CSS
- TanStack Router / Query (and related client data patterns in the web app)
- Privy authentication SDK and wagmi / viem for wallets
Data & infrastructure
- SQLite with Drizzle ORM for relational app state
- Redis for caching, sessions, job state, and real-time event streams behind SSE
- Qdrant vector database for course similarity and retrieval
- S3-compatible object storage and Synapse / Filecoin-oriented SDK usage for durable media where configured
AI
- TanStack AI with pluggable LLM and embedding providers (OpenAI-compatible and others as configured)
- Embeddings for course matching; LLM tasks for clarification, course authoring, page content, and explanations
- Sharp (and related tooling in the workspace) for image processing where needed
Blockchain
- Solidity smart contracts (Hardhat 3, OpenZeppelin contracts)
- viem for reads, writes, and typed ABIs
- Target chain: Flow EVM testnet configuration in the deployed setup
- Contracts include: orchestrator (XP mint + reward math), $PXP, $PANTHA (permit), shop, treasury, key store, certification authority, ERC-721 certificates
Quality
- Biome for formatting and lint in the workspace
- Bun test integration tests covering auth, courses, XP minting, streaks, social graph, encrypted chat, shop, and faucet flows
Pantha suits text-first subjects—history, sciences, art history, blockchain, languages—where reading, recall, and reasoning matter more than physical demonstration. It pairs that with crypto-native identity (wallet login), fair rewards tied to learning effort, and privacy-aware social features.