EMA

Inspiration

Traditional organizational climate tools act like a yearly photograph—by the time Human Resources receives the results, the data is already outdated and the underlying problems may have already escalated.

In high-stakes industries such as mining and field operations, delays in detecting burnout, operational friction, or psychological safety risks can affect both workforce wellbeing and operational continuity.

As psychologists with more than a decade of combined experience in organizational and clinical settings, we repeatedly observed the same problem: organizations often discover workforce issues only after they have become costly and difficult to address.

EMA was created to shift organizations from reactive measurement to continuous organizational intelligence. By combining behavioral science with Gemini, EMA helps leaders identify emerging workforce risks before they become operational problems.

What it does

EMA is an AI-native organizational intelligence platform designed for decentralized operations and frontline teams.

Through ultra-short anonymous pulse surveys (5 questions, approximately 30 seconds), sent on a regular weekly cadence, EMA continuously tracks workforce signals without creating survey fatigue.

Gemini analyzes aggregated trends, identifies behavioral patterns, detects emerging risks, and generates executive-level organizational insights while preserving employee anonymity.

Instead of delivering complex dashboards, EMA transforms workforce feedback into clear behavioral hypotheses and actionable recommendations that leaders can immediately understand and use.

How we built it

EMA was built entirely on Google Cloud, and we run the business itself the same way we build the product: AI-native, with a human founder in the approval loop.

Google Cloud Firestore stores aggregated organizational data while enforcing strict anonymity safeguards through application-level logic. Google Cloud Functions orchestrate survey collection, data processing, and report generation through an event-driven architecture.

At the core of the platform, Gemini 2.5 Flash on Vertex AI analyzes workforce dynamics using organizational psychology and behavioral science frameworks — CEAL-SM (Chile's official psychosocial risk instrument), the JD-R model, and Edmondson's psychological safety framework. Rather than generating generic summaries, the system produces structured organizational interpretations and decision-support recommendations for leaders, with confidence levels attached to every hypothesis so leaders know how much to trust each signal.

On the operations side, EMA's own engineering is executed by an AI coding agent (Google's Antigravity) that writes, tests, and proposes deploys directly against our Firebase/Cloud Run stack. The founding team acts as the governance layer: every production change requires explicit written approval and a reviewed diff before it ships, using staged preview channels to validate changes before they reach real users. This is deliberate — in a product handling psychosocial data for frontline workers, AI-native does not mean unsupervised, and building that discipline early is part of what makes the platform trustworthy enough to deploy in operational environments like mining.

Challenges we faced

Our biggest challenge was balancing analytical depth with absolute protection of employee anonymity.

To address this, we implemented aggregation thresholds, response minimums, and AI safeguards that prevent organizational diagnoses from being generated when there is insufficient data.

Another challenge was teaching Gemini to reason through organizational psychology frameworks instead of producing generic HR recommendations. Achieving this required extensive prompt engineering, testing, and iterative refinement.

A third challenge was operational trust: giving an AI agent real write access to our production codebase while working with sensitive workforce data. We addressed this with a strict human-approval protocol for every deploy, rather than letting the agent act autonomously in production.

What we learned

We learned that the most effective AI systems do not replace human judgment—they amplify it, both in the product we ship and in how we build it.

Leaders respond much faster to concise behavioral insights than to large volumes of raw data. We also observed that frontline workers provide more honest feedback when they trust that anonymity is genuinely protected.

Most importantly, we learned that organizations do not need more dashboards. They need better organizational understanding delivered at the right moment.

What's next for EMA

EMA is currently running its first commercial pilot within a mining organization (46 frontline and operational staff), validating the platform with real users under real operating conditions. The pilot closes in August 2026, with a commercial agreement and first payment in progress as the organization transitions from pilot to paid maintenance.

Our next objective is to increase the autonomy of the system by enabling Gemini to proactively identify workforce risks, generate richer organizational hypotheses, improve benchmarking capabilities across organizations in the same industry, and deliver increasingly sophisticated executive recommendations.

Our long-term vision is to create an AI-native organizational intelligence system that continuously helps organizations understand, anticipate, and respond to workforce dynamics before they become operational challenges.

Built With

• gemini-api
• google-cloud-firestore
• google-cloud-functions
• google-cloud-logging
• node.js
• typescript
• vertex-ia