This is a submission for the Presage SmartSpectra SDK Challenge. How do you know if a customer-service employee is ready to handle an angry or confrontational customer before they face one in real life?
Traditional customer-service training often focuses on what employees say, but it can overlook how they present themselves. A smile at the wrong moment, a visible frown, or an inappropriate response can make an already difficult conversation worse.
That's why I built Conflict Lab, a desktop customer-service readiness assessment tool powered by Presage SmartSpectra.
Conflict Lab places employees in realistic customer-service scenarios where a simulated customer becomes progressively more confrontational. As the employee responds, the application uses real camera-derived measurements to assess facial presentation, while AI evaluates the quality of their spoken responses.
The goal is to help employees identify areas for improvement before they face similar situations with real customers.
1. Personal calibration
Before the assessment begins, the employee completes camera and microphone checks, followed by personal calibration with neutral, smiling, frowning, angry-looking, and surprised facial expressions.
These serve as individual benchmarks, so Conflict Lab can compare facial observations against the employee's own references rather than a generic standard.
2. Realistic customer simulations
In the demo, the employee faces a simulated bank customer who disputes an unauthorized withdrawal.
The Full Assessment presents eight customer challenges across four escalation levels:
The employee responds verbally as the customer's attitude becomes more confrontational.
3. Two independent readiness assessments
Conflict Lab evaluates two areas:
4. Detailed assessment reports
After all eight exchanges, Conflict Lab produces a report with assessment scores, individual exchange evidence, facial observations, response feedback, and practical recommendations.
Employees can export reports as PDFs and save completed assessment records locally as JSON for later review and report regeneration.
Here's a real, complete eight-exchange assessment recorded directly from Conflict Lab.
Watch the Conflict Lab demo on YouTube The video shows the desktop application from camera setup and personal calibration through all eight customer exchanges, real Presage measurements, final assessment results, and report export.
| Measure | Result |
|---|---|
| Facial Presentation | 75/100, Grade C |
| Response Competence | 60/100 |
| Customer exchanges | 8 |
| Escalation levels | 4 |
| Personal facial calibrations | 5 |
| Qualified negative-looking facial event | 3.1 seconds |
The application recorded measurable changes throughout the simulation.
During exchange 4, the SDK's Happy classification rose 17.8 percentage points relative to the speaking reference.
During exchange 8, the Sad classification rose 32.2 percentage points relative to the speaking reference.
Conflict Lab also identified a sustained, calibrated negative-looking facial pattern lasting 3.1 seconds during the final exchange.
These are actual SDK classification changes and calibrated observations, not proof of the employee's internal emotional state.
GitHub repository: Conflict Lab Conflict Lab is built with Electron and JavaScript.
Its main components include:
The repository includes the application source, setup instructions, configuration examples, tests, and documentation of the assessment workflow.
Presage SmartSpectra is central to Conflict Lab. Without real camera measurements, the application cannot produce its facial-presentation assessment.
During the conversation, SmartSpectra provides timestamped facial-expression classification distributions across eight categories: Neutral, Happy, Sad, Angry, Surprise, Fear, Disgust, and Contempt.
As SmartSpectra provides new measurements, Conflict Lab updates its live facial-presentation display and retains valid samples for subsequent analysis.
The application distinguishes between the employee's listening and speaking phases, with separate personal references for each.
Instead of assuming everyone has the same neutral expression, Conflict Lab establishes personal calibration references.
The application compares subsequent observations against these references to identify meaningful changes in facial presentation.
This helps account for differences in how individuals naturally express themselves.
A brief facial movement should not automatically become a negative assessment.
Conflict Lab applies sample-quality checks, minimum durations, continuity requirements, and calibration comparisons before an observation qualifies as a scored event.
Qualified sustained facial patterns affect the facial-presentation score, while uncertain or short-lived observations remain unscored.
For example, the completed assessment recorded a strong negative-looking facial pattern lasting 3.1 seconds during exchange 8. The report retains supporting measurements and explains which observations qualified and which remained unclassified.
SmartSpectra also provides camera-derived pulse measurements when valid data is available.
Conflict Lab retains these measurements alongside facial-classification evidence, so employees can review measured physiological changes across different stages of a conversation.
The application connects the customer's statement, the employee's spoken response, Presage observations, and communication feedback within the same exchange.
This provides context for each result rather than a score without explanation.
The facial-presentation score and response-competence score remain independent. Groq evaluates spoken communication and does not invent or replace Presage measurements.
Conflict Lab does not claim to detect someone's true emotions, honesty, intent, or fitness for employment. Its scores are practice measures, not clinically validated judgments.
I'm entering both Best Thoughtful Real-World Use and Most Creative or Delightful Use.
Customer-facing employees regularly deal with complaints, accusations, and difficult conversations.
While they can practice scripted responses, they may not notice how their own facial presentation changes under pressure. Conflict Lab addresses this gap through voluntary, personalized assessment sessions.
It combines realistic customer interactions with actual Presage measurements and communication feedback, so employees can identify specific areas for improvement before they encounter similar situations at work.
The application preserves underlying measurement evidence instead of presenting unsupported conclusions about emotions or behavior.
What makes Conflict Lab different is the use of live human-signal technology within an interactive customer conflict simulation.
Instead of displaying facial classifications on a standalone dashboard, the application connects these observations to specific moments in an escalating conversation.
Employees can see how their facial presentation varies as the customer becomes more confrontational, then review the corresponding measurements and responses afterward.
This gives contactless human sensing a practical role in communication practice.
Conflict Lab is intended for voluntary employee practice sessions, not covert observation or automated employment decisions.
The application requires camera and microphone permissions before it can use those inputs. Users complete camera validation and personal calibration before the assessment.
Presage SmartSpectra provides camera-derived measurements, while Groq is an external service used for speech transcription and communication evaluation. Relevant audio and assessment text are transmitted to Groq when those services are used. Presage SDK processing may also involve network communication or provider telemetry according to its configuration and service policies.
Completed assessments can be saved locally as versioned JSON records and exported as PDF reports. These records include measurement evidence, calibration references, transcripts, and evaluation results. The JSON export excludes raw video, audio blobs, API credentials, and camera streams.
Local exports remain on the user's device unless the user chooses to share them. Users can delete exported files through their operating system. Provider-side retention and deletion remain subject to the applicable Presage and Groq service policies, and the application does not claim to control those provider-side records.
The current application does not claim automatic deletion of every locally retained or provider-processed record. Users should review the configured storage locations and relevant service policies before use.
Conflict Lab does not provide medical diagnoses, psychological assessments, deception detection, or automatic employment decisions. Its assessment scores are intended for practice and self-review.
I used OpenAI Codex to help build and test Conflict Lab, review the Presage integration, improve reliability, and develop the assessment and reporting workflows.