Case Study Β· Product #2 Β· In Development

Vansora Radio

A continuous, interruptible AI briefing stream.

Your daily information - news, work context, personal interests - delivered as audio. Interrupt to ask follow-ups. Resume with a recap. Every claim is source-backed or flagged as cautious. Trust and privacy by design.

179
Source files
11.2K
Lines of code
52
Test files
0
Vendor lock-in

Problem

The information problem

Too many sources. Not enough time. No way to know what's verified.

πŸ“°

Fragmented Sources

News tabs, Slack channels, email digests, RSS feeds, podcasts - context scattered across a dozen apps.

⏱️

No Time to Read

45 minutes before your first meeting, skimming headlines. Important context missed. Decisions made on half-information.

❓

Trust Vacuum

Which claims are sourced? Which are speculation? No system tells you. You guess, or you spend more time verifying.

Transformation

Measured transformation

Every improvement is quantified. Not promises - system capabilities.

Before

Skim 12 tabs of news, Slack, and email before 9am

After

One audio stream covers everything, prioritised by relevance

45 min β†’ 12 min morning prep

Before

No idea which claims are sourced vs speculative

After

Every statement colour-coded: verified or cautious

100% provenance visibility

Before

Interrupt a podcast - lose your place forever

After

Ask follow-ups mid-stream, resume with recap

Zero context loss

Before

Team briefings leak individual reading habits

After

Group topics shared, personal context partitioned

Cryptographic privacy boundaries

Before

Locked into one AI provider's quality and pricing

After

Swap LLM/STT/TTS adapters without code changes

Zero vendor lock-in

Capabilities

8 core capabilities

Not a feature list. An integrated audio intelligence system where every component reinforces trust, context, and continuity.

πŸŽ™οΈ

Continuous Briefing Stream

Daily audio briefings assembled from your chosen sources - news, work context, calendar, Telegram. Plays like radio, thinks like a research assistant.

⏸️

Interrupt & Resume

Pause mid-stream to ask follow-ups. The system answers in context, then resumes with a recap of what you missed. No lost threads.

πŸŽ›οΈ

Voice Commands

Pause, continue, skip, repeat, explain, go deeper, slower, faster, stop - a full command vocabulary for hands-free control.

πŸ”

Source-Backed Claims

Every statement is classified: verified (source-backed) or cautious (inferred). Green and amber chips make provenance visible at a glance.

🧠

Decision Traces

See why the system chose each segment, which sources were weighted, and what was excluded. Full reasoning transparency.

πŸ‘₯

Group Privacy Partitioning

Teams share briefing topics without leaking individual preferences. Cryptographic boundaries between personal and shared context.

πŸ”Œ

Provider-Agnostic Adapters

Swap LLM, STT, and TTS providers without code changes. OpenAI, Anthropic, ElevenLabs, Whisper - plug in what you trust.

πŸ“‘

Connectors & Delivery

RSS, Calendar (ICS), and Telegram connectors for source ingestion. Telegram delivery integration for on-the-go briefings.

Specs

Technical specifications

System Architecture

Source files179
Lines of code11,253
Test files52
DatabasePostgreSQL + Prisma ORM
StreamingSSE real-time delivery
DeploymentDocker + CI pipeline
ConnectorsRSS Β· Calendar (ICS) Β· Telegram
ValidationZod input schemas

Trust & Privacy

Claim enforcementSource-backed vs cautious
Decision tracesFull reasoning transparency
Group privacyCryptographic partitioning
Data retentionUser-controlled, deletable
Provider dataZero training on user content
AuthSession-based, encrypted

Briefing pipeline

sources.ingested β†’ classify_credibility β†’ rank_by_user_interest β†’ queue_segments
state: idle β†’ preparing β†’ fetch_sources β†’ generate_briefing β†’ ready
state: ready β†’ playing β†’ stream_audio_via_sse β†’ update_progress β†’ log_decision_trace
command: pause | skip | repeat | explain | go_deeper | slower | faster β†’ transition_state
state: playing β†’ awaiting_command β†’ process_follow_up β†’ generate_response β†’ resuming β†’ recap β†’ playing
session.completed β†’ generate_summary β†’ update_interest_model β†’ archive_traces β†’ ended

Interested in Vansora Radio?

We're building the next generation of information consumption. If you want early access or want to explore this for your team - let's talk.