FPI EVIT — eVIT Geospatial Platform

A geospatial desktop platform, plus agents that quietly give the team its day back.

For a specialist engineering firm we built a native C++ geospatial analysis application and a set of AI-agent automation flows — turning repetitive analysis prep and reporting into supervised machine work.

Project period

2023 — ongoing

Our role

Architecture and development of the desktop platform; design and operation of the agent automation layer (Python backend, TypeScript frontend).

Context

Geospatial analysis is expert work wrapped in clerical work: data wrangling, format conversion, report assembly. The experts were scarce; the clerical part ate their hours.

The costly problem

An estimated 100 person-hours per day across ~40 users went to repetitive preparation, conversion and reporting tasks — modeled from ticket logs and time sheets over 6 weeks.

Baseline

Ticket-log model validated with team leads: median 2.5 clerical hours per expert per day; error-prone manual re-entry between GIS, CAD and reporting tools.

Intervention

A native desktop application for the specialized analysis itself — fast, offline-capable, built for the actual workflows — surrounded by AI agents that prepare inputs, convert formats and draft reports for human approval.

Human responsibilities

  • Experts approve every agent-drafted report before release; nothing auto-sends
  • A power user per team acts as agent operator — first-line tuning, escalation to us second-line
  • Monthly adoption review with team leads decides what the agents take over next

Measured outcome

~60–100 h/dayperson-hours reclaimed across the team — modeled from logs, confirmed by leads
5.6k+lines of purpose-built C++ in the analysis engine
24/7agent availability with human checkpoints during work hours

Limitations

The 60–100 h/day range is a model, not a stopwatch measurement — inputs and assumptions are documented in the case artifacts and reviewable. Reclaimed capacity funded backlog reduction and training, not layoffs.

Experts spend mornings on analysis instead of preparation; report turnaround dropped from days to same-day. Adoption held because agents assist inside the tools people already use.

Next step

Extending agents into field-data intake and quality flagging.