Field Capture
A photo. An optional voice note. Submit. The system accepts what people already produce when they talk about a job, in the same form they already produce it. No new forms to fill out. No new behavior to learn.
Gregory Stoltz
Systems builder. Technology generalist. AI automation architect.
Greg helps small and mid-sized businesses replace paper, memory, scattered messages, and manual coordination with structured workflows, automation, and AI systems that remain understandable and controllable.
He has spent decades building systems that survive real operating conditions, from early Internet infrastructure to mobile field tools and AI-assisted workflows.
Operational reality
Photos in text messages. Voice notes between supervisors. PDFs nobody can search. Decisions buried in threads. The information already exists — it just walks out the door at the end of the shift.
The system collects those same inputs from where the work happens and builds a searchable record out of them. The technical name is a local-first multi-channel ingestion pipeline. The practical effect is that nothing useful goes missing.
A photo. An optional voice note. Submit. The system accepts what people already produce when they talk about a job, in the same form they already produce it. No new forms to fill out. No new behavior to learn.
OCR, transcription, tagging, summarization, embeddings, and retrieval layered onto raw operational artifacts.
Searchable records, summaries, dashboards, continuity, and organizational memory that survive turnover and chaos.
Commercial cleaning is the reference vertical. Distributed crews, multi-site routes, paper-and-text habits, and a low operational tolerance for software that gets in the way. The pipeline runs there in daily use. If it holds up in commercial cleaning, it holds up in adjacent field service work — HVAC service, pest control, landscaping, property management, route-based operations.
The pipeline runs on infrastructure I own and operate. The server is configured, paid for, and maintained by me. Inference — vision models for photo context, transcription for voice notes — runs locally, on hardware I control. There is no third-party service deciding what your operational data is worth, and no monthly bill that scales with how much of your reality you capture.
Turn scattered field notes, texts, spreadsheets, and memory into records people can search, audit, and use.
Use AI and scripts to reduce repetitive work while keeping the process understandable.
Design tools that survive spotty connectivity, messy handoffs, and real operating conditions.
The best systems do not start in conference rooms. They start with a real problem, under real pressure.
The current work begins by solving operational friction directly: field capture, shift notes, quality checks, site knowledge, supply signals, and daily reporting. Once a tool proves useful in live work, it can become a repeatable system. Once the pattern is proven across enough sites, it can become a model for the wider service industry.
That is the path: solve the real problem first, then let the system grow from evidence.
Live demo
The pipeline is live. FIN — the Feline Intelligence Network — is a public demonstration of the same ingestion and processing chain that runs in commercial cleaning operations: photo ingestion, transcription, tagging, semantic context, searchable records. Cats are the dataset because cats are mine to share. The architecture under it is the production system.
Submit a photo. Watch it move through the pipeline.
No cats are harmed. No data is used commercially. It is a working system.
Working systems matter more than demos, slide decks, and buzzwords. Good tools should make work clearer, not bury it under another layer of software.
A private operational intelligence system built around the reality your organization already produces every day.