Inspiration

Most sales funnels don’t die from lack of leads — they die in follow-up.
Leads come in hot, then hit objections (“too expensive”, “not sure”, “later”), and the conversation quality drops.

We wanted to build a closer that actually gets better with every interaction.


What it does

Adaptive Closer AI is a self-improving sales agent that:

  • handles inbound leads in text (Instagram DM / WhatsApp),
  • adapts strategy by objection type,
  • escalates high-friction leads to voice calls (ElevenLabs + Twilio),
  • ingests post-call data (duration + transcript signals),
  • and updates its policy for better next responses.

It’s not static automation. It’s a learning loop.


How we built it

We built a fast, reproducible pipeline:

  1. multi-strategy response generation,
  2. adaptive scoring (LLM-based with rule fallback),
  3. memory + policy updates over epochs,
  4. text-to-voice escalation orchestration,
  5. webhook ingestion with signature verification,
  6. judge-friendly one-command tests.

Everything is designed to be demo-safe and runnable in minutes.


Challenges we ran into

  • Shipping fast without losing architectural clarity.
  • Keeping the demo deterministic under hackathon pressure.
  • Balancing “smart LLM eval” with fallback reliability.
  • Adding security expectations (webhook signature verification) in a short time.

Accomplishments that we're proud of

  • Built a real self-improving loop, not just a scripted bot.
  • Added true multi-channel orchestration (text + voice).
  • Implemented LLM evaluator + robust fallback path.
  • Implemented production-style webhook verification.
  • Created copy-paste judge flows with observable proof markers.

What we learned

  • Adaptation beats raw automation.
  • Reliability wins hackathon demos.
  • Security and observability increase trust instantly.
  • Clear business narrative matters as much as code quality.

What's next for Adaptive Closer AI

  • Connect to real business inboxes at scale.
  • Learn from true conversion events, not only simulated rounds.
  • Upgrade to online bandit/RL-style policy updates.
  • Add a live dashboard for strategy and channel performance.
  • Launch pilots with SMB sales teams and optimize for measurable revenue lift.

Built With

Share this project:

Updates

Submission history