Data Loops (@data_loops) header
Data Loops (@data_loops) profile picture

CFO for 20 years. Now I build the tools I used to wish I had. Finance × Python × AI. Numbers must tie out.

Joined April 2015
Photo by Data Loops (@data_loops)Photo by Data Loops (@data_loops)Photo by Data Loops (@data_loops)Photo by Data Loops (@data_loops)Photo by Data Loops (@data_loops)Photo by Data Loops (@data_loops)Photo by Data Loops (@data_loops)Photo by Data Loops (@data_loops)Photo by Data Loops (@data_loops)Photo by Data Loops (@data_loops)
Everyone is talking about AI agents. AI is only one piece of the puzzle. I'm more interested in everything around them. Python, PostgreSQL, n8n, Docker, APIs, VPS. AI is only one piece of the puzzle. The real value comes from connecting everything together into systems that actually do useful work. That's what I'm building. I'll share everything I learn along the way. #AIAgents #Automation #Python
1
4
301
I didn't learn Python by taking a course. I learned it by turning my own morning routine into a script and asking "why" at every step: why a venv, why a config file, why the password isn't in it. The finance part I already knew. That turned out to be the hard part. #Python #FPandA #LearningToCode
2
28
The hard part of management reporting isn't the charts. It's proving that nothing fell out between the general ledger and the slide. So the controls run first: GL = trial balance, every account mapped, every manual adjustment with a reason and an author. Where AI helps: writing the SQL from the schema, building the account mapping, reconciling against the old manual reports. Where it doesn't: deciding what's right. The controls check the AI too. Stack: Database → Python → DuckDB → Report #FPandA #DuckDB #Python #DataEngineering #AI
4
45
🔁 Loop Check · AI 
Set an LLM's temperature to 0. Is the output guaranteed identical every run?
50%Yes
50%No
2 votes • Final results
1
2
66
✅ No. 
It becomes close to deterministic, but hardware and batching effects can still change results. Never build a control on "it gave the same answer last time".
6
🔁 Loop Check · SQL 
Column "amount" has values: 10, 20, NULL 
What does AVG(amount) return?
0%10
100%15
0%NULL
2 votes • Final results
1
1
105
✅ 15 
AVG ignores NULLs: (10 + 20)
10
A self-improving agent turns every task into a feedback loop. It executes, evaluates the result, learns from human corrections, stores useful knowledge, and applies it to the next run. Not autonomous evolution. Structured, measurable improvement. #AIAgents #AgenticAI #LLMOps #AIEngineering #Automation
1
80
One thing I underestimated: how much React speeds up internal AI tools. Instead of exposing raw APIs or SQL queries, I built a simple interface that let me search, inspect and validate registry data in seconds. Python handles the backend. PostgreSQL stores the data. React makes it usable. Good developer experience matters as much as good architecture. #React #Python #PostgreSQL #AI
1
2
62
One of the most satisfying moments in software development is when a project stops being “something running on your laptop” and becomes a real production application. Secure deployment. HTTPS. Frontend and backend working together. Database connected. Real users can access it. Building features is fun. Shipping them is even better. #buildinpublic #Python #React #FastAPI #PostgreSQL #Docker #WebDevelopment #SoftwareEngineering
1
50
Today wasn’t about building a new AI agent. It was about proving the platform can survive. Migrated the entire stack to a new VPS: • PostgreSQL (~18M+ registry records) • n8n • Nextcloud • Caddy Verified backups, restored services, validated data integrity, and brought everything back online. Infrastructure first. Agents second. Reliability always. #Python #PostgreSQL #Docker #n8n #DevOps #SelfHosted #AIAgents
1
1
76
Building AI agents isn’t just about prompting an LLM. Every useful agent needs a solid foundation: • reliable data pipelines • a clean business model • reusable services • automation that can run every day That’s what I’ve been building before adding more intelligence. A good AI platform isn’t a collection of isolated agents. It’s a system where every new agent becomes more useful because it builds on the same foundation. One step at a time. #AI #AIAgents #Python #PostgreSQL
1
4
87
People often ask why I’m using n8n if most of my platform is built in Python. Because each tool has a different job. Python handles the business logic. n8n orchestrates workflows, triggers, and integrations. That means I spend less time wiring systems together and more time building AI agents that actually solve problems. Use the right tool for the right job. #n8n #Python #Automation #AIAgents
6
87
A small decision that makes a big difference: Keep configuration and secrets in a .env file. Database credentials. API keys. Service URLs. Tokens. The code stays clean. Secrets stay out of GitHub. Each environment can use its own settings. Simple, but essential when you are building a platform instead of a collection of scripts. #Python #SoftwareArchitecture #DevOps #AIAgents
4
53
Core platform: done. ✓ Shared PostgreSQL database ✓ Docker environment ✓ Common Python framework ✓ API-ready architecture ✓ n8n integration Everything is now in place to start building production AI agents. #Python #PostgreSQL #Docker #AI
4
62
All my AI agents share the same PostgreSQL database because agents shouldn’t work in isolation. An email agent can collect data. A document agent can use it. A reporting agent can build on it. Small design decisions like this make it much easier to grow the system over time. #Python #PostgreSQL #AI
2
5
78
Today’s goal isn’t a smarter agent. It’s a better system. #DataLoops #AI
3
33
My first production AI agent has one job: Process incoming emails. Download invoices. Organize documents. Send a notification. That’s it. Just one reliable workflow that saves time every day. Build one useful agent. Then another. Eventually, they become a system. #AIAgents #Automation
4
78
Don’t put business logic inside n8n. Keep complex logic in Python. Let n8n orchestrate your workflows. This makes your automations easier to maintain, test and scale. How do you structure your n8n workflows? #n8n #Python #Automation
1
6
138
Stop deploying AI projects on your local machine. For less than $10/month, you can have your own Linux VPS running 24/7. My basic setup: • Ubuntu Server • Docker & Docker Compose • Nginx • PostgreSQL • Python • n8n • UFW Firewall • Automatic backups With this setup you can host: → AI agents → APIs → Telegram bots → Discord bots → Automation workflows → Personal projects Learning how to deploy your own server is one of the highest ROI skills for AI developers. It's much easier than most people think. What's stopping you from deploying your first VPS? #Linux #VPS #Docker #Python #AI
2
6
179