Storytime AI is a voice-first, multilingual storytelling platform for children aged 2–10. Built using Gemini, Whisper, LangChain, and ElevenLabs, it delivers personalized, screen-free narratives that adapt to each child’s voice input and language level. I led the product end-to-end — from concept to launch — optimizing for latency, education, and engagement.
Coming out soon on mobile and selected for Google’s AI Startup Program
Why We Built StoryTime
Parents told us:
- “Too much screen time.”
- “Apps don’t really listen.”
- “I need 10 minutes of peace.”
We built StoryTime to give children agency through voice and give parents peace of mind through privacy-first design.
How StoryTime Works
Parent Setup
Parent selects story length and theme preferences from our intuitive interface.
AI Greeting
TTS greets the child: “Hi Kiaan! What story would you like to hear?”
Voice Input
Child responds aloud. STT captures the input.
Story branches dynamically across 2–5 interaction points based on age and time.
Real-Time Adaptation
LLM uses token-based pacing to keep responses short, relevant, and engaging.
Satisfying Conclusion
Each session ends with a positive moral and a soft voice cue. No data is stored. Fully COPPA/GDPR compliant.
Engineered for latency <5s to match child attention span
User Personas & Journey
Primary User – "Kiaan", Age 4
Loves trains and animals, craves screen-free playtime with educational content. Enjoys voice-based interactions and simple prompts.
Secondary User – “Lucy”, Working Parent
Juggling work and parenting, seeks screen-free engagement for child, prioritizing learning and routine.
📖 Journey Overview
Before Storytime AI: Passive watching on YouTube. After Storytime AI: Personalized, interactive storytelling experience with metrics for improved engagement.
Customer Journey – From Curiosity to Connection
How parents and children experience StoryTime from first touch to final moral.
AI Hypothesis
If our conversational AI voice-agent engages children ages 3–6 in adaptive, research-backed language education through personalized, interactive storytelling—while maintaining at least 80% alignment with vocabulary objectives and a 100% child-safe content score—then we can create measurable learning gains and parent satisfaction, opening monetization opportunities through B2C (freemium/subscription) and B2B (early learning centers).
AI Architecture
Inputs: Child’s voice, name, language → transcribed by Whisper
Processing: Gemini 2.0 + LangChain (interaction logic)
Memory: MongoDB + LangSmith traces (session + preference)
Output: ElevenLabs voice (TTS) → personalized, multilingual
Loop: Real-time feedback & sentiment tracing refine future responses
*This is actively being updated and may not represent the current state (Moved from v1 - Universal and Open AI for STT and TTS)
Risks & Tradeoffs
What I Learned
- Designing for delight means designing for unpredictability.
Building for kids taught me to embrace ambiguity, not control it.
- Latency isn’t a tech issue — it’s a trust issue.
Speed = presence. Delays break not just flow, but emotional connection.
- Scope clarity is the most underrated skill.
Saying “no” to non-core features protected the product’s soul.
- Privacy isn't a feature — it’s a promise.
Earning parental trust required proactive, not reactive, guardrails.
Proven Results That Matter
Story Completion
Children consistently engage from start to finish
Pilot Participants
Successfully tested with families nationwide
Privacy Compliance
Full COPPA and GDPR-K certification
Selected by Google's prestigious AI Startup Program for innovation in child-safe AI technology. Our token-safe Gemini and LangChain implementation sets new industry standards.
Storytime AI Product Roadmap
Phase 1: MVP
- Voice-to-Voice MVP (English)
- Gemini 2.0 Flash integration
- LangChain for interaction handling
- Whisper for voice transcription
- ElevenLabs for output voice
Phase 2: Mobile & Multilingual Voice Support
- Mobile app launch
- Support for multiple character and language voices
- Offline mode
- Parental control for voice and language selection
Phase 3: RAG & Personalization
- Integrate RAG with LangChain
- Vector DB for reusable story components
- Support for multilingual user prompts
- Preference memory across sessions
- Checkpoint continuation logic
Phase 4: AI Feedback Loop
- LangSmith analytics traces
- Drop-off and interaction scoring
- Sentiment-driven fallback logic
- Adaptive prompt tuning and model refinement
Phase 5: Future Roadmap
- AI-driven voice style personalization
- Parent-facing dashboard and learning insights
- Story-level comprehension and engagement analytics
- Real-time multilingual narration powered by on-device inference
Customer Driven Product Strategy
Storytime's roadmap, architecture, and core features are grounded in real-world voice-of-customer insights. Each user story and feature set reflects tested needs from parents and children across multiple environments — from bedtime routines to multilingual homes.
The following sections illustrate how user input shaped our design, from MVP to multilingual, adaptive storytelling.
Key User Stories — Designed Around Real Use Cases
As a child, I want to speak to the app and hear a response in a fun voice,
so that I can enjoy a magical story experience hands-free.
As a parent, I want the story to wait for my child's response before continuing,
so that it feels more interactive and less like a passive video.
As a parent, I want to select different story voices and language preferences,
so that my child can hear familiar or diverse voices and learn new languages.
As a parent, I want to quickly open the app and see recent stories and themes played,
so I can track what my child is engaging with.
As a child, I want the story to remember my favorite animal and my name,
so that it feels like it was made just for me.
As a parent, I want the story to help teach empathy or problem-solving,
so my child can learn while being entertained.
As a parent, I want to see how long my child engaged and where they dropped off,
so that I can understand attention span and improvement over time.
As a PM, I want to know which interaction points are working and which themes are skipped,
so that I can fine-tune our AI prompts for better outcomes.
As a multilingual parent, I want stories to be available in multiple languages,
so that my child can learn and connect in our native language.
As a parent, I want to ensure the story is safe, educational, and ad-free,
so I feel confident letting my child use it independently.
These stories directly shaped Storytime's product architecture, roadmap phases, and interaction point logic._
Success Metrics & AI-Specific KPIs
Voice Interaction Metrics
- Completion Rate: % of stories finished in one session
- Interaction Accuracy: STT-to-expected intent match rate
- Drop-off Points: Avg. story step where users disengage
- STT Latency: Avg. Whisper response time in seconds
Personalization Impact
- Re-engagement Rate: % of users returning within 3 days
- Story Continuation Rate: % of stories picked up across sessions
- Vector Hit Accuracy: RAG success in reusing correct segments
- Preference Recall: # of personalized variables retrieved/session
AI Feedback & Prompt Loop
- Prompt Adaptation Score: % of successful fallbacks triggered by sentiment
- LangSmith Trace Depth: Avg. steps traced per interaction path
- Fallback Trigger Rate: % of sessions needing recovery logic
Parental Engagement & Trust
- Voice Customization Usage: % of parents customizing voice
- Language Preference Setting: % of sessions using non-default language
Product Requirements & Acceptance Criteria
MVP Requirements
Interactive Story Generator, Voice-to-Voice Interface, Theme Selection Engine, Parent Dashboard
Post-MVP Requirements
Multilingual Story Support, Sentiment-Adaptive Storytelling, Memory-Based Personalization
Global Safety & Quality Criteria
COPPA/GDPR compliance, audio moderation, token limits, age-appropriate interaction times
- Stories must have branching paths and pass content filters
- Voice narration must be responsive and accurate
- Themes must influence story content and be trackable
- Parent dashboard must be mobile-friendly and data-driven
- Multilingual support, sentiment analysis, and personalization
- Strict safety and quality standards for child users
📘 Storytime AI Case Study
Principal Product Manager (Founder-led)
Led the full lifecycle — from concept → MVP → mobile platform with multilingual, voice-first experience. Defined architecture, roadmap, and metrics for production-quality AI integration.
Team: 4 PMs, 2 frontend devs, 1 ML engineer, 1 content designer
Stack: Gemini 2.0 Flash, LangChain, Whisper STT, ElevenLabs TTS, LangSmith, MongoDB, Vector DB
Tools: Notion, Figma, Retool, ElevenLabs, Latitude AI, Replit, Claude, DoveTail
- Built MVP in 60 days → 100% early user retention
- Selected for Google’s AI Startup Program
- Reduced parent interaction time by 40%
- RAG loop with LangSmith + fallback tuning
- Personalization for multilingual homes
"From bedtime stories to multilingual, intelligent narration — Storytime is voice-first AI storytelling that listens, adapts, and delights."
Used by families across 3 continents. Built for kids aged 2–10. Designed for high recall, repeat engagement, and voice-based learning.
🎖️ Designed to demonstrate Principal PM-level thinking, execution, and AI fluency across Gemini, LangChain, and RAG loops.
Think I might be a good fit for your team?
📩 Lets connect 🔗 LinkedIn
Storytime AI: Voice-First, Mobile, and Built for Imagination