Introduction: The Problem & The Vision
Navigating academic roadmaps, entrance exams, admissions, and financial aid can be an overwhelming journey for students across India. Text-heavy portals, complex application forms, and confusing documentation guides often create barriers rather than bridges.
To bridge this gap, I built EduGuideAI—an interactive, ultra-low-latency voice assistant designed to guide students and parents through their educational queries naturally. Built during the 10 Days of Voice Agents — #VoiceForBharat Challenge by Murf AI, EduGuideAI moves beyond traditional chatbots by holding real-time, code-mixed (Hinglish/English) spoken conversations, executing live database checks, remembering student context, and escalating complex inquiries to human counselors.
How EduGuideAI Works (System Architecture)
A smooth voice experience requires near-instantaneous streaming across four primary stages:
Audio Transport: Real-time WebRTC audio handling via LiveKit Agents. #
Speech Recognition (STT): Captures accents, terms, and conversational student phrasing. #
Reasoning & Tool Execution: LLM loaded with academic guardrails, student profile memory, and functional tools. #
Voice Synthesis (TTS): Murf Falcon — delivering lightning-fast streaming audio with natural, clear Indian pronunciation.
Key Features Built into EduGuideAI
- Instant Indian Voice with Murf Falcon
Academic guidance requires an empathetic, reassuring, and articulate tone. With Murf Falcon, response latency stays under 800ms, making conversations feel natural and fluid without awkward s.
- Guardrails & Code-Mixed (Hinglish) Support Students often mix Hindi and English when asking questions (e.g., "Is college ke admission criteria kya hain?"). EduGuideAI naturally parses code-mixed queries while staying strictly within safe academic boundaries, preventing hallucinations on critical details like deadlines and fee structures.
- Persistent Memory & Real-Time Student Tools
Dynamic Information Retrieval: Fetches live details regarding courses, deadlines, and documentation checklists mid-dialogue. #
Student Context Memory: Remembers returning students' previous queries, target fields of study, and application stages.
- Outbound Telephony & Counselor Escalation
Outbound Updates: Initiates reminder calls for upcoming application deadlines or documentation verification. #
Human Counselor Handoff: Cleanly routes sensitive counseling or specialized career discussions to human advisors along with the conversation summary.
Engineering Challenges & Solutions
Handling Backchannel Cues & Interruptions:
Problem: Students often interject with acknowledgments like "haan", "samjha", or "okay" while the bot is explaining a process. #
Solution: Fine-tuned Voice Activity Detection (VAD) thresholds in LiveKit to distinguish casual backchannel listening cues from true user interruptions.
Mitigating Latency During Database Queries:
Problem: Awaiting API responses for course or cutoff lists created silent gaps. #
Solution: Configured EduGuideAI to emit conversational filler phrases (e.g., "Checking the latest guidelines for you...") while asynchronously fetching tool data.
How to Run EduGuideAI Locally
Inspect the complete codebase on GitHub:
👉 murf-livekit-starter Repository
- Clone the Repository