Skip to content
Code by Sergio
PTENES
  • Virtron
  • Conversational BI
  • In production

SalesBot

Ask about the operation.Get the answer straight from the CRM.

Internal projectProprietary code

Back to category
Sales Bot

The problem

The data was already in the CRM.There just wasn’t a way to talk to it.

Every quick lookup meant opening Ploomes, filtering and interpreting by hand. The Sales Bot moved those questions to Telegram.

How it works

One source, two layers.

  1. Straight from the CRM

    Notifications and questions query the Ploomes API, with no spreadsheet in between.

    1. 01Ploomes API
    2. 02Python
    3. 03Telegram

  2. Parallel layer

    The ETL that feeds the executive dashboard and historical analysis.

    1. 01Ploomes
    2. 02Python ETL
    3. 03Google Sheets
    4. 04Looker Studio

    In Google Sheets

    1. SDR
    2. Vendas
    3. Cohort
    4. Cohort Long

Another one!!

The team and the board stay up to date without opening the dashboard.

  1. Detection within ~5 minutes
  2. No duplicate notifications
  3. Consolidated daily close
New-sale notification on Telegram

Conversational BI

Questions in Portuguese, by text or by voice.

  • “How many sales did we have today?”
  • “What about yesterday?”
  • “And by salesperson?”
  • “What was this month’s average ticket?”

The AI understands the question.The system finds the answer.

The LLM only steps in when the parser can’t resolve a question — and only to interpret it. Querying Ploomes, filtering and calculating stay with the code, to reduce hallucination.

Interpret

  1. 01Question
  2. 02Deterministic parser
  3. 03LLM fallbackOnly when needed
  4. 04Structured intent

Calculate

  1. 05Ploomes API
  2. 06Filters & aggregation in Python
  3. 07Deterministic answer

  • 0 tokens

    Simple questions resolved by the deterministic parser

  • 15 min

    Of context for follow-ups

  • ~60 s

    Voice messages via Telegram, transcribed locally with Whisper

What you can ask

Metrics

  • Number of sales
  • Revenue
  • Average ticket
  • R$/kWp
  • New leads
  • Lost deals

Filters & groupings

  • Salesperson
  • City
  • Source
  • Payment method
  • Bank
  • Period

Including rankings and period-over-period comparisons.

Impact

  • No manual export or merging

  • Dashboard that updates itself

  • Sales notified on Telegram

  • Lookups without opening the CRM

Tech specs

Backend / Data
PythonPandas
CRM
Ploomes API
AI / NLP
Claude HaikuWhisper
Interface
Telegram Bot API
Dashboard
Google SheetsLooker Studio
Infrastructure
Linux

Internal projectProprietary codeIn production, with parts still evolving