All work
Client work through InsideBoard
Automating client-file checks
Showed it could remove more than ten hours of manual file checks a week.
I worked with InsideBoard’s teams to map the existing data flow, challenge the approach, then design a file-intake solution inside their platform and build its proof of concept with Claude. InsideBoard’s engineering team then took the platform over to scale it.
Forward Deployed Engineer, InsideBoard AI · May 2026 to present · Enterprise SaaS, automotive data
A closer look
InsideBoard automation25 sec25 sec · silent
Recorded workflowRecorded client workflow
The problem
InsideBoard receives recurring Excel files from a large automotive company. Their structure can change without warning, so two people had to inspect each file before processing could continue. The client needed proof that an AI agent or an automated solution could handle them.
The outcome
The two people who checked the files by hand tried the proof of concept: the files went straight through, without manual checking. InsideBoard’s engineering team then took the platform over to scale it. People step in only when a file needs attention, and every run is traceable.
My contribution
I gathered input from the people responsible for the process, platform and operations, then mapped the flow from file arrival to delivery. I defined the solution architecture and chose where n8n orchestration, Python validation and AI analysis belonged. AI advises on file structure; deterministic checks decide whether delivery is allowed. Coding agents wrote the code under my direction; I reviewed, tested and evaluated each step, including the connections between SFTP, the validation service and the Slack alerts, within InsideBoard’s GitLab delivery workflow. The proof of concept showed the approach works, and InsideBoard’s engineering team then took the platform over to scale it. The recording shown comes from the development environment.
The specifics
- Business and technical inputs brought together into one intake-to-delivery design
- Clear boundaries between workflow orchestration, AI advice and deterministic validation
- Versioned template rules and explicit exception handling, so changing files do not silently pass through
- Connected implementation with tests, operator alerts and traceability for each run
How it works
- Each file follows the same path: identify its template, read it, check it, get AI advice on its structure, apply deterministic checks, normalise it, deliver it
- AI advises on file structure only, never on the cell data. Deterministic checks decide whether a file is delivered
- A file that no longer matches its template is stopped and flagged, never passed through silently
- Operators follow each run live in a console, see the health of every file template, and get a Slack alert when a file needs attention
- A versioned API (20+ endpoints) runs the jobs in the background, with every AI call traced
Safeguards and quality
- Every change is tested automatically before release (550+ tests), with code-quality checks and a review on each merge request
- Direct pushes to the main branch and any leak of secrets are blocked automatically
- Architecture, operations and security decisions are written down, so the team can see why the system works the way it does
- Third-party AI tools are checked for prompt injection before they are allowed near the project
Stack
- Python
- FastAPI
- n8n
- openpyxl
- LiteLLM / Anthropic
- Langfuse
- PostgreSQL
- GitLab CI/CD
- Slack