Inspiration

For independent contractors and small trade businesses, a phone line is both a lifeline and a liability. Inbound calls are a chaotic mix of high-value homeowners, general contractors needing bid compliance packets, and aggressive telemarketers.

Unsolicited lead-generation solicitors don't just call once—they run aggressive predictive dialers hitting a line 4 to 6 times in rapid succession. Instead of losing hours to phone screening or letting valid estimates slip away, CallSmith AI turns inbound telephony into an autonomous triage, lead-conversion, and legal defense engine.

What It Does

CallSmith AI acts as an autonomous front-desk commander that classifies every incoming call in real time and executes dynamic operational branches:

  • Homeowner Lead Qualification & VIP Fast-Track: Detects residential project inquiries, logs customer scope, and automatically sends an SMS asking for project photos. When the homeowner replies with images, CallSmith AI elevates the lead directly to a dedicated #vip-leads Discord channel for priority estimating.
  • General Contractor (GC) Compliance Fast-Path: Identifies commercial project inquiries and general contractors instantly. The agent clarifies required vendor paperwork, collects the estimator's email address, and queues an automated compliance packet dispatch (Certificate of Insurance / COI, W-9, and trade licenses).
  • Spam Stalling & Dynamic DNC Enforcement: Intercepts spam solicitors and deploys Gemini to hold convincing, dynamic conversations—wasting the telemarketer's call-center bandwidth instead of yours.
  • Repeat Offender Evidence Locker (TCPA Case Builder): If the caller ignores Do Not Call (DNC) status or calls back across rapid dialing bursts, CallSmith records audio transcripts, timestamps, and caller identities. It automatically compiles a structured evidence dossier for statutory TCPA claims ($500 to $1,500 per willful violation).

How We Built It

  • Intelligence & Orchestration: Built with the Google GenAI SDK using Gemini for real-time intent classification, context steering, and transcript-to-JSON parsing.
  • Telephony & Anti-Spoofing: Powered by Twilio with cryptographic STIR/SHAKEN verification to detect and neutralize local neighbor spoofing before calls reach the honeypot pipeline.
  • Voice Pipeline: Integrated with Google Cloud Text-to-Speech API to deliver low-latency bidirectional conversational responses.
  • Backend Infrastructure: Packaged in Python and containerized for Google Cloud Run, handling asynchronous webhook intake, schema validation, and event dispatching.

Challenges We Ran Into

  • Structuring dynamic multi-path branching that switches instantly between cordial customer intake, professional GC onboarding, and adversarial telemarketer stalling.
  • Normalizing messy conversational audio into strictly typed JSON schemas to ensure evidence dossiers meet legal logging standards.

Accomplishments That We're Proud Of

  • Turned an inbound spam liability into an automated asset: high-priority residential leads get fast-tracked, commercial compliance friction is eliminated, and repeat telemarketers build their own legal paper trail with zero manual effort.

What's Next for CallSmith AI

  • Adding automated small-claims demand packet generation and direct integration with municipal permit lookup APIs.

  • Are you a trade contractor or field business tired of robocalls and missed estimates? Join the closed pilot waitlist: https://forms.gle/J7FmGQPydcQF8r8y9

Built With

  • docker
  • fastapi
  • gemini-3.6
  • google-cloud-run
  • google-cloud-text-to-speech
  • google-genai-sdk
  • python
  • webhooks
Share this project:

Updates

posted an update

Since submission, I've made the lead generation route to be more robust to detect lying solictors, trying to get passed Gemini, with a caller confidence score. I've also shortened gemini's response time even further. Its now deployed on both my person and business lines!

Log in or sign up for Devpost to join the conversation.

Submission history