Built a native Text-to-SQL AI agent from scratch using Python and Docker.
No LangChain or visual wrappersโjust raw tool-calling, deterministic read-only guardrails, and a self-healing execution loop.
Demo & repo: https://t.co/9gSs0MIql3
https://t.co/dM42RFoR9c
Built a Text-to-SQL AI agent in raw Python (zero LangChain).
๐ Guardrails: Blocks malicious UPDATE/DROP prompts.
๐ Self-healing: Fixes its own SQL errors.
#Python#AI#BuildInPublic
Basic chatbots are out. Bespoke AI agents are in. ๐
Just built a self-hosted Support Agent with n8n, Pinecone, and a local LLM via FastAPI.
The best part? Zero hallucinations. It refused to make up a fake phone number and safely escalated the ticket instead. ๐
#AI#n8n
๐ง Hardest lessons learned:
โข Windows CMD mangles JSON payloads (use PowerShell Invoke-RestMethod!)
โข n8n wraps incoming webhook data in a body object (remap those expressions).
โข Local hosting requires tight ngrok tunnel and Uvicorn routing management.
#AI#Automation#n8n
โ๏ธ The Flow:
1๏ธโฃ Webhook catches unstructured lead text 24/7
2๏ธโฃ Local Llama 3.1 extracts Buyer Intent, Budget & Timeline
3๏ธโฃ n8n appends the structured data to a Google Sheets CRM
4๏ธโฃ "Hot" leads trigger an instant Telegram push notification to my phone ๐ฅ
๐ Just built a fully autonomous, self-hosted AI Lead Processing Engine!
No SaaS lock-in. Just raw engineering.
๐ ๏ธ Stack: n8n + Llama 3.1 + FastAPI + Google Sheets + Telegram
๐ Here is how it works and what I learned building it:
Day 2: Connected my self-hosted AI lead engine to Google Sheets & Gmail โก
1๏ธโฃ Fixed LLM JSON extraction in FastAPI
2๏ธโฃ Mapped data straight to CRM
3๏ธโฃ Set conditional n8n routing for instant ๐ฅ HOT lead email alerts!
Pipeline is live. ๐
#BuildInPublic#AIAutomation
Day 1: Building a self-hosted Real Estate AI Lead Engine ๐งต
Architected state tracking CRM logic & built a Python @FastAPI + Pydantic brain (Llama 3.1).
Connected local @n8n_io โ ngrok โ FastAPI to parse raw SMS into clean JSON in seconds โก
#AIAutomation#BuildInPublic
Just deployed my first end-to-end AI web app! ๐
I built an Autonomous PDF Invoice Extractor to automate manual accounting data entry from unstructured docs.
Stack: OpenRouter API, PyMuPDF, Pydantic, Pandas, & Streamlit.
#AI#Python#Automation
@naval Absolutely, technology amplifies both collaboration and opportunities. Embrace the leverage and break the barriers. The winners inspire growth for all!