Nafiou Tino

0 likes • Jan 25

@Waled Ahmdi With pleasure! 🤝

2 likes • Jan 25

@M. talha Faraz 🤝🤝

Julian Blazhev

Jan 18 •

Best way to build RAG

Hey everyone, I am building a RAG for a lawyer company (working with ton of documents). What do you think is the best way to build RAG (Lang Chain, Lang Graph, Supabase, etc.)

3 likes • Jan 21

For legal specifically, the bar is much higher. You can't afford hallucinations or missed citations. My recommended stack for this: 1. Database: Supabase (pgvector). This is crucial because it supports Hybrid Search (Keyword + Semantic). Lawyers often search for specific terms (e.g., 'Section 4.2'), which pure vector search sometimes misses. 2. Orchestration: LangGraph. Standard LangChain is linear. For legal, you need an agentic loop (e.g., Retrieve -> Grade Relevance -> If bad, Rewrite Query -> Retrieve again). LangGraph handles this logic perfectly. 3. The 'Secret' Sauce: Parsing. Do not use standard PDF loaders. Legal docs have columns and tables that break standard chunking. Look into LlamaParse or Unstructured to clean the data before embedding it. Garbage in, garbage out is 10x truer in Legal Tech.

0 likes • Jan 21

@Julian Blazhev 🤝

Yousef Arebi

Jan 18 •

Need help

If I have a customer who needs an AI agent for customer support, which option should I use: n8n plans or a VPS? I see that the n8n plans have limits on the number of executions you can have. Which one should I go for?

0 likes • Jan 21

use vps

Kavish Sheth

Jan 20 •

n8n Security

Hey guys! I have built 3 interlinked n8n workflows for one of my clients. All of them are triggered by webhooks. In order to deploy them, I purchased a separate n8n instance for the client. What can I do in the workflows to ensure data security for the client? Please comment if you need more information to give me a better response. Thanks!

0 likes • Jan 21

Great move isolating the instance! That’s 90% of the battle. For the workflows themselves, since they are webhook-based, your main goal is Authentication and Validation. 1. Lock the Webhooks: In the Webhook node settings, don't leave it open. Enable 'Authentication' (I prefer 'Header Auth'). Create a custom secret/API key (e.g., x-api-key) that the sender must include. If they don't, n8n rejects the call immediately. 2. Restrict HTTP Methods: Set them to POST only. Never pass sensitive data via query parameters in a GET request (as those often show up in server logs). 3. Validate Data: Use the built-in Schema Validation in the webhook node (or a logic node right after) to ensure the incoming JSON is exactly what you expect. This prevents bad actors from injecting malicious code. Simple but effective!

Karim Janoun

Jan 20 •

For those running AI receptionists / voice agents:

Which automation flow has been the most reliable in production for you? I’m optimizing for low failure rate, clean automations, and scalability — curious what’s actually working in the real world, not theory.

For those running AI receptionists / voice agents:

Level 4 - Agent Builder 🤖

0 likes • Jan 21

vapi is much better.

1-10 of 13

46points to level up