Auricle

Features

• Real-time call routing and audio handling across the VoIP telephony stack (SIP registration, PSTN connectivity, RTP media processing)

• Live captioning pipeline routing call audio to speech-to-text processing and streaming captions back to the app in real time, with failover across multiple providers

• Data platform syncing user, billing, and verification records to internal systems and data lakes via automated ETL pipelines

• Push notification and voicemail dispatch for call and account events

• Centralized monitoring and debugging tooling across the full call-relay pipeline

Challenges

• Voicemail system (Vodia) failing to scale with user growth, driving churn and falling app store ratings as captioning reliability degraded

• Hard-deadline data pipeline migration required to meet the TRS-URD compliance transition, no flexibility on timing

• Executive-backed push to migrate to a new captioning provider for six-figure monthly savings, despite vendor benchmarks not capturing real usability gaps

• Captioning servers had no load balancing across the pool; an overloaded instance had no failover path and would degrade until it lagged out and crashed, causing outages during peak usage hours

• Customer support and operations pushed for a fully automated backend endpoint to replace the existing verification/scrubbing workflow, competing for engineering time against the captioning server reliability issue

Key Decisions

• Prioritized the TRS-URD data migration first, since it had a hard regulatory deadline with no flexibility on timing, ahead of two other high-pressure initiatives, while sequencing the voicemail (Vodia → Twilio) migration second based on measurable and worsening impact, active churn and declining app store ratings, over less time-sensitive work

• Pushed the AI captioning provider migration to later despite executive and AI-department pressure and six-figure monthly savings, running internal user studies first since vendor benchmarks didn't capture real usability issues

• Deprioritized building a fully automated backend endpoint for the verification workflow in favor of the captioning server load-balancing fix, since the existing process was functional, if less scalable, and not causing active harm, while the load-balancing gap was actively causing outages

Impact

• 98%+ of active users remediated to TRS-URD compliance and migrated to URD endpoint ahead of the FCC's hard deadline

• Avoided shipping a degraded captioning experience for hard-of-hearing users by running internal usability studies

• Improved SLA reliability by eliminating peak-usage outages in the captioning pipeline by resolving the load-balancing gap across the server pool