Software · Working application
Local AI Marketing & Coding Agent
A local Ollama-powered Streamlit workspace with product knowledge, strategy, content, outreach, SEO, campaign, coding, and library tools.
Problem
A solo founder needed consistent product knowledge, content, outreach, SEO, and code generation without sending every workflow to a cloud service or rebuilding context each time.
Description
What I built or investigated
A local Streamlit application using Ollama with a maintained product profile and separate workspaces for strategy, ideas, targets, content, images, outreach, SEO, campaigns, coding, and a library.
Iterations
The work did not happen in one jump.
- 01
Created a product brain as the source of truth.
- 02
Added task-specific prompt functions instead of one generic chatbot box.
- 03
Tuned GPU, context, batch, and model settings for local hardware.
- 04
Added strict plain-English output rules after coder models produced corrupted marketing text.
- 05
Separated coding requests from marketing requests so each could use a more appropriate model.
Result
Where the project landed
A working local founder-operations tool with saved outputs and a clear expansion path.
Lessons
What carried forward
- Model choice matters more than model size when the task type changes.
- A maintained context object is more reliable than repeatedly explaining the business.
- Local tools still need validation, error handling, and UX boundaries.
Images & evidence
Work in context
Media still being recovered
This case-study location is ready for photos, diagrams, source screenshots, or video when they are available.
Documentation status
This is a living case study. New photos, measurements, diagrams, source files, and exact outcomes can be added without changing the permanent project URL.
Continue exploring
Every catalog entry has a permanent page.
Use the archive to move between embedded systems, software, research, fabrication, networking, manufacturing tools, and field repairs.