Customer Service · Tabnews · 🇧🇷 Brazil
Stopping your WhatsApp agent from duplicating customers in the CRM
A post shows why agents write to the wrong record: accents, invisible characters and stylized names.
- Profession
- Developer, CRM Manager
- Tools
- Python, LLMs
- Source
- tabnews.com.br ↗
- Country
- 🇧🇷 Brazil
- Report date
- October 24, 2026
- Verified on
- October 24, 2026
What they did
Tested text normalization pitfalls between the conversation and the database.
How it went
The author says that in most cases they've seen, the error is in the bridge to the database, not in the model. The account gives no numbers.
Good to know
Merging the wrong records mixes personal data of different people: an incident under LGPD.
The prompt
Write a function in [language] that normalizes names and phone numbers (NFC/NFKC, removal of invisible characters, area code) only to SEARCH for possible duplicates in the CRM, without merging automatically. Include tests with "João" written with accents in different ways.
Replace what is in [brackets]. Prompt written by us from the account, not a literal copy of the source. Do not paste other people’s personal data (privacy laws such as LGPD and GDPR apply).
Summary in our own words. Results as stated in the original account; we have not tested or audited them. Read the source ↗ Found a mistake?
Similar cases
01Adding an assistant that answers only from your WordPress site's content
A dev published a plugin with RAG and Gemini in the official directory that prevents made-up answers.
02Building an AI WhatsApp bot for small businesses
A developer built a customer service SaaS with Next.js and Baileys to avoid the official API's fees.
03Clinic customer service on WhatsApp without the AI making up prices
A developer used RAG so the bot answers only with the clinic's real data.
04Generating all of a game's pixel art with code written together with AI
A dev gave up on image generators and started drawing isometric sprites with a Python script.
05Building an AI system for psychologists from the countryside
A 20-year-old in Tobias Barreto (SE) created EmotionAI, AI-powered clinical management for psychologists.
06Build prospecting lists from Google Maps
A developer built a local lead extractor with Python and Playwright, speeding up development with AI.
07Following the AI blindly and breaking Linux (and learning from it)
A developer describes how commands suggested by an AI broke their dual boot and what they learned.
08Use a local AI to detect phishing
A security professional showed a Python script using a local model to analyze the intent of suspicious emails.