The client is private about what we're building together, so this case study describes the work without naming them.
The client
A specialized technical services company whose experts diagnose and resolve complex problems in the field. The people using the platform are deep subject-matter experts, not software or AI specialists.
The challenge
- Every hard problem the team solved produced valuable knowledge: notes, photos, screenshots, instrument data, and the eventual resolution. That knowledge was scattered across devices, message threads, and individual memories.
- When a similar problem came up again, there was no reliable way to answer "have we seen this before?" Expertise left with each job and each departing employee.
- Relevant external reference data existed, but there was too much of it. Experts had to dig through hundreds of loosely related records to find the few that mattered.
- Specialized data files needed dedicated desktop software to open, which made them hard to share or review on the go.
- Any AI had to be trustworthy. For experts making consequential decisions, an AI that confidently makes things up is worse than no AI.
What we built
A custom, multi-tenant web platform for phones and desktops that serves as the team's system of record and shared memory. It runs on DigitalOcean.
Fast capture from the field
- Start a new job record from a phone in seconds, with key identifying details filled in automatically.
- Structured write-ups with templates for common job types, auto-save, and a domain-aware spellchecker.
- Attach photos, PDFs, and screenshots. The Google Cloud Vision API extracts the text and recognizes what's in each image, so even a photo becomes searchable.
- Turn attachments into presentation-ready slides and export them to PDF or PowerPoint.
Specialized data, viewable anywhere
- Technical data files that normally require desktop software open in an interactive browser viewer, including on mobile.
- Proprietary file formats are converted automatically in the cloud.
The right reference data at the right moment
- Several government and other public data sources feed the platform, some streamed on demand and some imported daily.
- For each job, the few records that match the problem rise to the top, each with a plain-language reason it matched. The full set stays one scroll away.
- Inconsistent naming in the external sources had caused some records to be silently missed. We fixed that at the source.
"Have we seen this before?" with AI the team can trust
- The platform surfaces the team's own prior jobs that resemble the current one, ranked and previewable in place.
- An AI assistant summarizes what those prior jobs suggest: the likely direction, next steps, and cautions. Users can then ask follow-up questions.
- Built-in honesty: the assistant can only reference real records the team has, every reference is checked before it's shown, and it says so plainly when the evidence is thin.
- Each organization's data stays private to that organization.
Built for real-world reliability
- Long-running AI tasks continue in the background and survive page refreshes.
- Automatic recovery from temporary outages at outside services.
- Per-organization feature switches for gradual rollout.
- Usage and cost dashboards keep AI spend visible and predictable.
- Leadership reporting on team documentation activity.
Our approach
- Built alongside working experts, with frequent releases and direct user feedback.
- AI that supports expert judgment instead of replacing it: show the evidence, explain the reasoning, and be honest about uncertainty.
- Fix problems at the source instead of working around them.
- Treat reliability and cost control as features from day one.
Outcomes
- One searchable home for every job and everything attached to it.
- Institutional knowledge that compounds: each documented job makes the next similar one faster.
- Relevant reference data surfaced in seconds instead of buried in noise.
- AI assistance grounded in the team's own experience, with safeguards against made-up answers.
Capabilities demonstrated
Custom web application, multi-tenant SaaS, applied AI and grounded assistants, knowledge management, document and media management, OCR and image recognition, external data integration, interactive data visualization, mobile-first UX, background processing, and cloud cost monitoring.
Related services: custom software, AI & LLM, and integrations.