Experience
DIKI GmbH
- I'm building a (not-so-controversial) Palantir for PR agencies.
- My AI agents collect and structure messy market data so agencies can find clients they didn't know existed. Conversion rates up 20x for the companies using it.
- DIKI also changes cold outreach. Agencies upload a target list, DIKI generates an audit report on each target's communication weaknesses and they send it before the first call. Clients are impressed.
- €5k MRR. Currently bootstrapping.
- I taught myself to vibe-code by building this, got better as I went, working with clients and studying at LSE at the same time.
Gallery
Operator flow for a DIKI run. Agencies upload a target list (CSV or Excel), preview rows and map columns to company fields, set prompts and analysis options, choose output format or destination and which AI model to use, then start processing. This is where messy spreadsheets become a structured pipeline toward intel and audit-style outputs.
Gated admin view (restricted to configured admin emails). Summary of total orders and completed vs waiting jobs, number of user accounts, and usage analytics for successfully generated PDFs: time range filters, daily volume bar chart, and split of activity for billing and health checks.
Server-side operations screen: free space on the project volume, count of live manual registry jobs (separate from always-on daemons), and status of background services raul_crawler and raul_cleaner with PIDs, scheduler state, and stop controls. This is part of how market data is collected and cleaned continuously in production.