SALES-BOT

One-liner: A sales negotiation co-pilot for wedding vendors — import a call transcript and get a report card on what you said, what the client signaled, and what to try next.

Origin (the Build Loop)

Because I sell wedding services directly to couples and lead every sales conversation myself, I kept hitting the fact that I’d hang up the phone and immediately forget half of what the client signaled — the objections, the budget tells, the things they laughed at, so I built a post-call analyst that reads transcripts and scores the conversation against pricing psychology patterns, that gives me a report card and a growing playbook of what works with real couples. It taught me that the most valuable LLM application isn’t generating text — it’s analyzing your own pattern against a human domain you thought you understood.

The problem

Sales coaching for wedding vendors doesn’t exist — there’s no “go practice your consultation” simulator and no way to debrief a call systematically. Every conversation with a prospective couple has tells: how they react to a price, what they laugh at, where they pause. A human salesperson can feel “good” or “bad” about a call but can’t score it objectively. I wanted a tool that would read a transcript and show me exactly where I left money on the table.

What it does

• Transcript analyzer — paste in a sales call transcript and get a structured report card on rapport, objection handling, value articulation, and close technique.
• Pricing psychology patterns — detects where the conversation touched on anchoring, decoy effects, loss aversion, and social proof.
• Growing playbook — every analyzed call contributes to an evolving guide of “what works with this demographic” based on real outcomes.
• Upsell engine — surfaces margin opportunities by reading what the client signaled they valued (extra hours, specific services, premium add-ons).
• Rule-based scoring without an API key — core analysis works offline; LLM-powered event classification is optional via OpenRouter.

How it’s built

• Stack: Python, FastAPI, SQLite, HTMX, dark-theme dashboard; OpenRouter (optional LLM layer)
• Notable engineering:
  • Rule-based + LLM hybrid — core pricing-psychology analysis runs on deterministic rules so it works without any API key. The LLM layer (OpenRouter) enriches it with sentiment and event classification when configured.
  • Domain-specific scoring — the rules encode real sales psychology (anchoring, decoy, loss aversion), not generic sentiment analysis.
  • Lightweight deployment — runs on port 5559, single binary-equivalent Python app, SQLite-backed.
• Architecture: FastAPI serves the HTMX dashboard; transcript text in → rule engine → optional LLM enrichment → report card → playbook update.

Proof points

• Live and running — served on a port-ready instance, taking real wedding transcript analysis.
• Dual-mode analysis — works fully with rules alone, improves with LLM enrichment when the key is set.
• Domain-specific knowledge encoded — the playbook grows from real calls, not generic sales training.

What to show

• Demo: The dashboard at the configured port showing a sample transcript report card and the growing playbook.
• Visuals needed: the report card view showing scores per category; the playbook page showing accumulated patterns; a sample transcript annotated by the tool.

Cross-links

• Built as a WeddingOS companion — SALES-BOT closes the loop from lead inquiry to signed contract with better margin.
• Shares the domain-specific analysis DNA with POLYBOT (where the domain is betting markets, not sales calls).

Case-study angle

This is a perfect example of Alex building for his own daily workflow: a wedding-vendor salesperson who built his own post-call coach because the training doesn’t exist elsewhere. The rule+LLM hybrid means it works regardless of budget, and the playbook grows from real Indianapolis wedding leads. It’s the domain-moat principle applied to sales.