Internal AI Tools
Your team googles, copy-pastes, asks around. And loses hours doing it. I build you a tool that finds knowledge, understands documents and gives answers. Not a prototype nobody uses. Something that runs in everyday work.
AI agents and process automation from Stuttgart, built by Brandon Agil
You have a process that's annoying you. Too much manual work, too little overview, something always takes forever. I build the system that fixes it. AI, automation, whatever works. No agency, no bullshit.
2 open days
Based on my calendar.
10 open slots over the next 5 workdays.
What I Build
First I understand your problem. Then I build. Not the other way round.
Automate processes
The same steps, every time: update the CRM, pull a report, reconcile data. No human needs to do that. I build the bridge between your tools and take the busywork off your plate.
AI agents
Not the 100th chatbot. Agents that actually get things done: following up leads, prepping data, qualifying customers. Your team does less monkey work, more real work.
From idea to live
I don't just build the AI part and hand over a concept paper. Backend, frontend, deployment, all from one hand. In the end the thing is live and you're working with it.
This is what it looks like
ACISO x ELEMENTS Fitness
Keeping members before they leave
The problem: gyms only notice a member is about to churn once the cancellation is already in. Too late.
What I built: a system that predicts churn before it happens. ML scoring per member, an agent that automatically suggests actions, and a dashboard so the team knows where to look. No more manual clicking through lists. The system does the work, the team takes the action.
- Lead time
- 30 days before a member could cancel
- Members scored
- 12k+ daily, fully automated
PythonOpenAI APIBigQuery MLNext.jsCRM
Running for me
Agentic RAG server
Genome Intelligence
An agent I can talk to about my complete genome data. Millions of variants, dozens of data sources, all on my own machine.
The agent has over 30 tools and picks the ones it needs for each question itself: looking up genes, checking drug interactions, pulling studies from the GWAS Catalog. It doesn't dump the whole dataset into the model, it only fetches the rows it needs for the question. That keeps answers fast and costs low. Every statement comes with a source, without proof it won't answer. SMBs who want an agent over their own documents need the exact same pattern.
- Agent tools
- 30
- Data sources
- 11
- Cloud uploads
- 0
FastAPISQLite + DuckDBSSE streamingTailscale-privateOn-device
Local speech pipeline
Applaud
Recording in, finished protocol out. Transcript, recognized speakers, summary and action items, entirely on the local machine.
I took an open-source meeting-recording tool and built it out into my own pipeline: Whisper transcribes, Pyannote separates the speakers, a match against voice profiles names them, a model cleans up the text and pulls out the tasks. A webhook pushes the protocol to n8n automatically. On top of that I built my own MCP server: it lets my AI assistant search the entire meeting archive and answer questions about it, without a single transcript ever leaving the machine. For businesses whose conversations have no business in someone else's cloud, that's the requirement, not the feature.
- Processing
- local
- Speaker ID
- yes
- Audio in cloud
- never
WhisperPyannoteSpeaker IDMCP servern8n webhooks
How agents work
Demo for sales
Inquiry Agent
Inquiries land in the inbox and wait. Whoever replies after three days has often already lost the job.
The agent reads every email and every contact form, pulls out quantity, date and material, checks the price list and capacity, and drafts a ready-to-send reply. Spam and job applications get filtered out right away. A human still approves it, nothing goes out unchecked.
- Draft reply
- 41 sec
- Manual before
- 2.4 hrs
- Pre-qualified
- 96%
Email inboxLead scoringPrice list checkApproval workflow
Demo for accounting
Invoice Agent
Every invoice gets typed up, coded and filed by hand. Early payment discounts expire because nobody's watching.
The agent reads every invoice, extracts all fields, runs the calculation check, screens for duplicates and suggests the account coding. Fully coded, it goes straight into the DATEV export. When something's unclear, it asks instead of guessing.
- Per invoice
- 38 sec
- Manual before
- 9 min
- Auto-processed
- 87%
Invoice extractionDATEV exportDiscount watcherFour-eyes principle
Demo for customer service
Support Agent
Same questions, every day. Where's my package, how does a return work. And the customer still waits hours for a reply.
The agent answers standard cases on its own. For everything else it drafts a reply, with sources from your own material: manuals, FAQs, product range. Every statement is backed up. Whatever it isn't sure about, it escalates to your team.
- Solved autonomously
- 58%
- Response time
- 4 min
- Manual before
- 5.1 hrs
Knowledge base (RAG)Shop integrationSource citationsEscalation rules
Experience
Mercedes-Benz AG
Product Management MB.OS
Evaluated AI use cases for the vehicle operating system. What makes sense, what's hype. Translated between tech teams and business, built market analyses, set priorities. Learned: the best ideas fail on bad communication.
MHP / Porsche
Working Student Software Engineering
Internal tools and client projects at the Porsche subsidiary. The job where I realized I'd rather build than consult.
About Me
Short version
Business informatics background, currently in a Master's in Entrepreneurship. Saw at Mercedes-Benz and MHP/Porsche how corporates talk about AI, and how little of it actually gets built. So now I do it myself: build systems that run. Not slides that look good.
My Stack
- Python
- TypeScript
- Next.js
- FastAPI
- OpenAI API
- LangChain
- Docker
- BigQuery
FAQ
How does working together look?
Intro call, 30 minutes, free. We look at where you're losing time and whether AI or automation actually helps. If yes: a small, clearly scoped first step that goes live fast. Then we iterate. No spec-document marathon, no workshop theater.
Who do you work with?
Teams who want problems solved, not concepts sold: SMBs, studios, agencies, individual departments in larger companies. Happy to meet in person around Stuttgart, otherwise remote across Germany.
What's different from an agency?
You talk directly to the person building it. No handovers between consulting, design and development, no overhead padding the invoice. When something breaks, you call the person who wrote the code.
What tech do you work with?
Python, TypeScript, Next.js, FastAPI, OpenAI API, LangChain, Docker, BigQuery. More important than the stack: I integrate into what you already have (CRM, databases, internal tools) instead of forcing a new system on you.
How do I get started?
Book an intro call at beagil.de/booking or send an email to brandon@beagil.de. A short description of the problem is enough, we'll figure out the rest in the call.