All work
InsideBoard AI
Automating client-file checks
A deployed system that checks incoming Excel files, prepares the usable data, and brings the team in only when something is wrong.
Forward Deployed Engineer, InsideBoard AI · May 2026 to present · Enterprise SaaS, automotive data
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 intervention
I designed and delivered an automated intake system that checks each file against the expected structure and prepares it for the next step. If a file is not compliant, the team receives the precise reason and can act before any incorrect data is delivered. Files arrive over SFTP, pass through registry-proven Python validation and normalization, and are delivered back over SFTP, the full flow orchestrated by n8n.
The outcome
The automated checks save the team more than ten hours each week. People now step in only when a file needs attention, and every run remains traceable.
The specifics
- Manual line-by-line file review replaced by an automated, proven pipeline
- Operator target: any file outcome resolved in under 60 seconds
- Versioned registry for high-column-count Excel templates; the recorded template contains 89 expected columns
- Every run traced and permanently logged, AI reasoning attached to each flag
How it works
- 11-step Python intake pipeline: fingerprint, parse, prove, AI analysis, deterministic belt, normalize, deliver
- Guardrail model: the LLM advises on structure only (never cell data); a V1-V10 integrity belt and Python proof decide delivery
- Self-improving agent memory: observations promote to lessons only past a proof threshold or human approval
- FIP operator console: SSE-streamed live pipeline, template-health governance, multi-tenant client portals (EN / FR)
- Versioned REST API (20+ endpoints), background job runners, Langfuse tracing, Slack operator alerts
Why you can trust it
- Multi-agent dev harness: tool-scoped review agents (architecture, operational, security, planning) that vote on every change in sequence, plus a knowledge-graph code index agents query instead of grepping raw source
- Tiered context system: a controlled-vocabulary glossary, numbered session-handoff docs, and per-agent persistent memory with automated nightly consolidation, so any fresh AI session rebuilds full project state without re-reading the codebase
- Append-only decision log (60+ locked ADRs), plus a security gate that vets every third-party agent or skill for prompt injection before it touches disk
- Four-stage CI: 550+ hermetic tests with coverage, SonarQube, per-PR preview builds, and an automated AI code-review gate, behind push-to-main and secret-exfiltration guard hooks
Stack
- Python
- FastAPI
- n8n
- openpyxl
- LiteLLM / Anthropic
- Langfuse
- PostgreSQL
- Kubernetes
- Argo CD
- Slack
A closer look
InsideBoard automation25 sec25 sec · silent
Recorded workflowRecorded client workflow
Have a project like this?