Real-Time Voice Conversation Engine
OpenAI Realtime API powers natural, low-latency voice conversations - speech in, contextual response out, within seconds.
LifeWritr - Turning Spoken Memories into Written Legacies
Most people will never write their life story - not because it doesn't matter, but because writing is hard and the blank page is intimidating. LifeWritr solves this with an AI-powered platform that captures life stories through natural voice conversations and transforms spoken memories into structured, written autobiographies families can treasure forever.
This case study covers how Brainspack built LifeWritr - from real-time voice AI and intelligent interview orchestration to narrative structuring and a sharing system that lets loved ones explore the finished story.
The Problem
Why Life Stories Go Untold
The client had a deeply personal mission: make it easy for anyone to preserve their life story before it’s too late. They’d seen too many families lose irreplaceable memories when a parent or grandparent passed away - stories that existed only in someone’s head, never written down, never recorded, gone forever. The idea was simple: what if you could just talk, and AI would do the rest?
But simple ideas hide enormous complexity. The client had already explored basic transcription tools - record a conversation, get a transcript. The result was unusable. Raw transcripts are disorganized, repetitive, full of tangents, and read nothing like a coherent narrative. What they needed wasn’t transcription. They needed an AI system that could interview someone like a skilled biographer, organize memories into a meaningful narrative arc, handle contradictions and timeline gaps gracefully, and produce something families would actually want to read.
They came to Brainspack looking for an AI development company that could build the full product - voice AI, intelligent interviewing, narrative structuring, and a polished mobile experience - from the ground up.
The Challenges
Talking Is Easy, Writing Is Hard
Raw spoken memories are messy - the AI needed to extract meaning and organize them into structured chapters, themes, and timelines.
An AI Interviewer, Not a Transcription Tool
The AI needed to actively guide conversations - asking follow-ups, probing gaps, and responding sensitively to emotional moments.
Life Stories Are Messy and Contradictory
People remember out of order - contradictions, gaps, mixed timelines - all without disrupting the natural flow of storytelling.
Real-Time Voice with Low Latency
Sub-second response times and a natural-sounding voice - fast enough that users never felt like they were talking to a machine.
Cross-Platform Mobile Experience
One button, start talking - the interface had to be that simple, especially for older adults less comfortable with technology, across both iOS and Android
Privacy-Sensitive Content at Scale
Life stories are deeply personal - encrypted storage, strict access controls, and storyteller-controlled sharing were non-negotiable.
Technologies Used
Python
FastAPI
AWS
Prisma
GPT-4o
Building an AI Biographer That Listens, Remembers, and Writes
We approached LifeWritr as three interconnected systems: a Voice AI Layer that handles the real-time conversational experience, an Intelligence Layer that manages the interview strategy, memory retrieval, and narrative arc, and a Structuring Layer that transforms raw spoken content into organized, readable chapters. All of it wrapped in a mobile app simple enough for someone’s grandmother to use.
Intelligent Interview Orchestration
The Interview Orchestrator tracks covered periods, fills gaps, probes emotional moments, and picks up exactly where the last session left off.
Memory Retrieval & Conflict Resolution
A searchable memory bank surfaces past memories naturally, flags contradictions silently, and lets users resolve them during review sessions.
Narrative Structuring & Generation
Raw transcripts are transformed into polished, chapter-organized autobiographies - exportable as Legacy Letters or Just-in-Case documents.
Sharing & Access Control
The storyteller controls exactly who sees what - invitation-based access with granular permissions, chapter-level visibility, and fully revocable access for every reader.
Cross-Platform Mobile App
React Native app with a one-tap recording flow - plus scheduled phone dial-in sessions for users who prefer traditional calls.
Measurable Impact
LifeWritr’s success is measured in stories preserved and families connected. Here’s what the platform achieved:
< 2s
Voice Response Latency
27
Database Entities Across 10 Domains
45 min
Avg. Session Duration
1-Tap
One-Tap Recording Flow
Quality Outcomes
Frictionless Onboarding
First voice session completed within 2 minutes of download - one-tap flow eliminated adoption friction for older users.
Stories That Read Like Books
Chapter-organized autobiographies that users described as feeling like a real book - not a transcript.
Multi-Session Continuity
Full context maintained across 10+ sessions - memory retrieval built naturally on every past conversation without users repeating themselves.
Family Engagement
Family members actively explored shared autobiographies - with full storyteller control over access, and Just-in-Case documents particularly valued for end-of-life planning.
Accessible to All Ages
Phone dial-in expanded access to elderly users who prefer traditional calls - same AI quality, no smartphone required.
Product Discovery & Research
Defined LifeWritr and LifeReader personas, mapped the full user journey, and prioritized simplicity above everything.
System Architecture
Designed two-service architecture and a 27-entity database schema across 10 domains - users, interviews, documents, sharing, and more.
Voice AI & Interview Engine
Integrated OpenAI Realtime API and built the Interview Orchestrator with life period tracking, gap analysis, and adaptive questioning.
How We Built LifeWritr
Memory & Knowledge Base
Built semantic memory retrieval using Qdrant and PostgreSQL, with session context assembly and a Conflict Resolution Engine.
LifeWritr Delivered
Sub-2-second voice response, 45-minute average sessions, and a one-tap experience completed within 2 minutes - cross-platform, zero friction.
Build & Deployment
Built the React Native app, invitation-based sharing, Stripe billing, phone dial-in, and deployed on AWS ECS with encrypted storage.
Narrative Generation
Built the AI pipeline transforming raw transcripts into polished, chapter-organized Legacy Letters and Just-in-Case documents.
What Makes LifeWritr Unique
Voice-First, Zero-Typing Experience
One tap, start talking - the AI interviews, transcribes, and structures the full narrative. No typing, no buttons, no friction.
AI That Interviews Like a Biographer
The Interview Orchestrator tracks covered periods, identifies gaps, asks follow-ups, and knows when to go deeper - like a skilled listener who never forgets a word.
Life Period Coverage Tracking
Visual life coverage tracking balances user-led storytelling with gentle nudges toward gaps - ensuring a comprehensive autobiography, not just favourite memories.
Automatic Conflict Detection
When contradictions arise, the AI flags them without interrupting the conversation - users resolve conflicts during review sessions, keeping the final narrative accurate.
Two Document Types for Different Needs
Two document types - Legacy Letters (full autobiographies) and Just-in-Case documents (focused messages for specific people) - auto-generated and fully editable.
Phone Dial-In for Non-App Users
For users who prefer traditional calls, LifeWritr schedules phone sessions - same AI interviewer, same memory system, same narrative output, no smartphone required.
I came with a vision and uncertainty about whether it was possible. Brainspack LLP showed me how it would work, then built it exactly that way. The voice experience feels natural, the stories are genuinely beautiful, and my mother - who can barely use email - completed three sessions on her own. That's the ultimate test, and they passed it.
LifeWritr
Founder
Project Screenshots
Common Questions About This Project
LifeWritr is an AI-powered voice autobiography platform. Users open the mobile app and have natural voice conversations with an AI interviewer that guides them through their life story. The system records, transcribes, and organizes those conversations into structured, readable autobiographies - complete with chapters, timelines, and narrative prose. Family members can then be invited to read and explore the finished story.
Transcription gives you a messy, unorganized dump of speech. LifeWritr does five things transcription can’t: it actively interviews you (asking follow-up questions, guiding the conversation), it remembers everything across multiple sessions, it tracks which life periods you’ve covered, it resolves contradictions in your timeline, and it generates polished, chapter-organized narrative prose that reads like a real autobiography.
LifeWritr uses OpenAI’s Realtime API for bidirectional voice streaming (the user speaks and the AI responds with natural speech), Deepgram for high-accuracy real-time transcription, and GPT-4o for interview intelligence, follow-up question generation, and narrative prose writing. The entire voice pipeline runs with under 2-second latency.
That was a core design requirement. The app interface is extremely simple - open the app, tap one button, start talking. For users who don’t want to use a smartphone at all, LifeWritr also supports phone dial-in sessions where the system calls the user at a scheduled time and conducts the interview over a regular phone call.
Life stories contain deeply personal information, so LifeWritr was built with healthcare-grade security. All data is encrypted at rest (AWS S3) and in transit. The storyteller controls exactly who can read their story through an invitation-based sharing system with granular permissions. Access is revocable at any time.
LifeWritr uses a two-service architecture: NestJS for the application backend (user management, billing, sharing) and Python/FastAPI for AI workloads (voice processing, interview orchestration, narrative generation). The mobile app is built with React Native for iOS and Android. The database layer includes PostgreSQL with Prisma ORM (27 tables, 10 domains), Qdrant for memory search, and Redis for caching. Deployed on AWS with Docker containerization.
Yes. The voice AI architecture behind LifeWritr - real-time conversation engines, intelligent interview orchestration, memory systems, content structuring - is adaptable across use cases. Whether you need a voice-based customer onboarding experience, an AI interviewer for research, a voice-first knowledge capture tool, or any application where speaking is easier than typing, we can build it.
Have an idea for a voice-powered AI product, or need to build an intelligent app from concept to launch?
We build full-product AI applications - from voice AI and conversation engines to mobile apps and cloud deployment. If your product needs real-time voice, intelligent content generation, or AI that adapts to each user, we’d love to hear about it.