EdexConvoAI — The Adaptive Voice Tutor for Real Learning

EdexConvoAI is a real-time, voice-native AI tutoring platform that adapts to every student through Socratic dialogue, Bloom's Taxonomy progression, cross-session memory, and four distinct learning modes — all powered by Agora Conversational AI Engine.

 

---

 

3️⃣ Problem Being Solved

Traditional EdTech is broken in a fundamental way: it delivers information but doesn't build understanding. Students watch videos passively, read static text, or answer MCQs with instant hints — none of which trains the deep retrieval and explanation skills required in exams, vivas, or real jobs.

The core problem:

• Passive learning doesn't stick. Students need to explain, not just recognise.

• One-size-fits-all content ignores where each student actually is cognitively.

• No continuity — each session starts from scratch, wasting time.

• Viva and oral exam preparation has no scalable solution.

• Study groups have no AI facilitator that can probe, challenge, and adapt.

EdexConvoAI solves this with a voice-first, adaptive, memory-aware AI tutor named Lexi that meets every student exactly where they are.

 

---

 

4️⃣ Target User

Primary market size: 1.5 billion students globally; 50M+ preparing for oral/viva examinations annually.

 

---

 

5️⃣ How Agora Conversational AI Is Used

EdexConvoAI is built natively on Agora's Conversational AI Engine — Agora is not just transport; it is the intelligence layer:

The system prompt (lib/edex-prompt.ts) is built dynamically from the learner's profile and chosen mode — injected at agent invite time via the Agora Agents API.

 

---

 

6️⃣ Demonstrated Conversational AI Capabilities

 

---

 

7️⃣ External Action by the Agent

"At least one external action or structured outcome" — ✅ Multiple:

1. Session Report Card — After every session, Lexi's conversation produces a structured JSON-grade report (strengths, weak areas, recommended revision topics, performance grade A–F). This report is exportable via clipboard and persisted to localStorage for the teacher.

2. Cross-session memory write — The session summary is serialised into the learner profile (previousSessionSummary, previousSessionMistakes) and injected into the next session's system prompt via the Agora Agents API — creating a genuine memory loop without a backend.

3. Dynamic system prompt construction — /api/invite-agent builds a 2,000+ character adaptive prompt from live learner state (Bloom level, mastery scores, past mistakes, chosen mode) and calls the Agora Agents API to start an agent session with that context.

 

---

 

8️⃣ List of External APIs, LLMs, Speech Providers & Services

No third-party LLM API key required. The LLM, STT, and TTS are all managed by Agora's AI Engine.

 

---

 

9️⃣ Known Technical Limitations

1. No persistent backend — Learner profiles and session memory are stored in localStorage only. Data is lost if the browser cache is cleared or a different device is used.

2. Mastery scoring is heuristic — Topic mastery is estimated from conversation patterns (keyword detection), not from a formal assessment engine.

3. Single student per session — The current architecture supports one-to-one tutor/student. Multi-student collaborative sessions (study groups) are not yet implemented.

4. Offline unavailable — Requires an active Agora network connection; no offline mode.

5. Language — Currently English only; multilingual mode is architecturally planned but not wired.

6. Session length — Very long sessions (>45 min) may hit Agora's managed agent timeout; the app handles this gracefully via the ErrConflict 400 pattern.

7. Next.js 16.2.6 stale version warning — Cosmetic; does not affect functionality.

 

---

 

🔟 AI Limitations & Safety Considerations

 

---

 

1️⃣1️⃣ System Architecture

┌─────────────────────────────────────────────────────────────────────┐

│                         BROWSER (Next.js Client)                    │

│                                                                     │

│  ┌──────────────────┐    ┌──────────────────────────────────────┐  │

│  │  EdexOnboardingCard│   │        ConversationComponent          │  │

│  │  (4 learning modes,│   │  ┌─────────────┐  ┌──────────────┐  │  │

│  │  subject, Bloom,   │   │  │LexiAvatarPanel│  │TranscriptPanel│  │  │

│  │  analogy domain)  │   │  │(orbital rings,│  │(live turns,  │  │  │

│  └─────────┬─────────┘   │  │ state glow)  │  │ agent state) │  │  │

│            │             │  └──────┬───────┘  └──────────────┘  │  │

│            ▼             │         │                              │  │

│  ┌──────────────────┐    │  ┌──────▼──────────────────────────┐  │  │

│  │  LearnerProfile  │    │  │ AgoraVoiceAI (toolkit)          │  │  │

│  │  localStorage    │    │  │ RTM: TRANSCRIPT_UPDATED          │  │  │

│  │  - Bloom level   │    │  │      AGENT_STATE_CHANGED         │  │  │

│  │  - masteryScores │    │  │      AGENT_METRICS               │  │  │

│  │  - mistakes      │    │  └──────┬──────────────────────────┘  │  │

│  │  - past summary  │    │         │                              │  │

│  └──────────────────┘    └─────────┼────────────────────────────┘  │

│                                    │ Agora RTC + RTM               │

└────────────────────────────────────┼────────────────────────────────┘

                                    │

┌────────────────────────────────────┼────────────────────────────────┐

│                    NEXT.JS API ROUTES (Server)                      │

│                                    │                               │

│  /api/generate-agora-token  ←──────┤  RtcTokenBuilder.buildTokenWithRtm │

│  /api/invite-agent          ←──────┤  Agora Agents SDK (managed session)│

│    └─ buildLexiPrompt()            │  2000+ char adaptive prompt    │

│  /api/stop-conversation     ←──────┤  Agora Agents API              │

└────────────────────────────────────┼────────────────────────────────┘

                                    │

┌────────────────────────────────────┼────────────────────────────────┐

│              AGORA CONVERSATIONAL AI ENGINE                         │

│                                    │                               │

│   [STT] → Whisper / Agora ASR  ────┤                               │

│   [LLM] → Managed LLM (prompt) ────┤  VAD: start_of_speech /       │

│   [TTS] → Agora managed TTS    ────┘       end_of_speech           │

└─────────────────────────────────────────────────────────────────────┘

 

---

 

1️⃣2️⃣ Future Evolution

 

---