• 

  App_Logo

Argo — AI Phone Assistant (Hackathon Story)

Inspiration

Waiting on hold to fix bills is soul-sucking. Argo lets users clone their voice and delegate calls to an AI “personal assistant” that navigates IVRs, talks to reps, and summarizes outcomes—while keeping a friendly, trustworthy brand (royal blue, off-white, golden accent; diagonal-phone+robot icon).

What I learned

• How to encode product taste in prompts (XML alpha prompts for UI, voice, and agent etiquette).
• Designing an iOS app so backend swaps are painless (clean view→intent→state boundaries).
• Telephony UX matters: consent, transparency, and readable post-call notes beat raw automation.

How we built it

• Frontend: SwiftUI iOS app (local MVP), screens for Voice Clone, Call Setup, Live Status, and Outcomes.
• Backend (local MVP): modules for user auth, voice-clone I/O, call control, LLM agent, and call summarization; designed as swappable services (telephony, TTS/ASR).
• Brand & assets: iterative logo/icon work (headset assistant inside a diagonal phone) using Argo’s palette for a modern, credible feel.

Savings model (for dashboards): [ S = \sum_{i=1}^{n}\big(b_i - \tilde{b}i\big),\quad \text{where } b_i=\text{original bill},\ \tilde{b}*i=\text{negotiated bill}. ] Success across steps (reach agent, verify, negotiate): [ P{\text{success}} = 1-\prod*{k}(1-p_k). ]

Challenges

• Latency & barge-in: timing TTS/ASR with IVR menus.
• Provider variability: dynamic prompts for unpredictable scripts.
• Safety/consent: clear user authorization and voice-use disclosure.
• Summaries that matter: compress a 20-min call into a 5-second brief.

Outcome

A working iOS MVP scaffold with branded assets, prompt packs, and a local backend prototype that already produces usable call summaries—positioned to plug into hosted telephony and scale.

Built With

• chatterbox
• elevenlabs
• gemini
• swift
• twilio