Nabd is a web system supporting preliminary psychological assessment of war-affected children by analyzing non-verbal behavior and voice with Gemini, live or from video.
Nabd is an intelligent web system that supports the preliminary psychological assessment of war-affected children by analyzing non-verbal behavior and vocal cues with artificial intelligence, developed as an academic project by a team of students with technical assistance from Techno Enjaz. It lets a therapist run a live analysis session using the camera and microphone or analyze a recorded video; the data is sent to the Google Gemini 3 Flash Preview model, and the system displays numeric indicators from 0 to 100 and behavioral labels in Arabic, then helps prepare a preliminary report that the therapist reviews and approves.
The system is built with a React 19 and TypeScript front end, a Laravel 12 server, and a MySQL database. In testing within a simulated environment, it recorded a response time of 1.5 to 4 seconds per analysis frame, with one frame captured every 3 seconds. The project stresses that Nabd is a decision-support tool, not a replacement for clinical diagnosis; the decision remains with the psychologist.
| Item | Details |
|---|---|
| Project type | AI-supported web platform for psychological assessment |
| Field | Multimodal AI, child mental health, medical informatics |
| Project status | Working system tested in a simulated environment for live-analysis and video-analysis sessions |
| Techno Enjaz role | Technical support and assistance to the students during project development |
| Core technologies | Google Gemini 3 Flash Preview, React 19, TypeScript 5.8, Vite 6, Tailwind CSS 4, Laravel 12, PHP 8.2, Axios, MySQL |
| Outputs | Live and video analysis, numeric indicators and Arabic labels, real-time alerts, child profiles, reports with review and approval |
War and armed conflict leave long-lasting psychological effects on children, such as anxiety, depression, and post-traumatic stress disorder (PTSD), and these conditions are hard to detect early because many children cannot accurately put their feelings into words.
The project describes the limits of traditional assessment tools in unstable settings: structured clinical interviews need a highly trained interviewer, a lot of time, and a quiet setting; self-report questionnaires depend on the child's ability to understand and express themselves; and parent questionnaires may be affected by the parents' own trauma. Add to this the shortage of specialists, language and cultural barriers, social stigma, and the fact that these tools give a one-off snapshot rather than continuous monitoring. The project therefore turned to non-verbal behavior and voice as indicators that do not depend on language and can be captured with simple devices.
The project was carried out academically by a team of students, with technical assistance and support from Techno Enjaz during development. It is presented among the office's work as an example of its support for projects that apply multimodal models to sensitive health applications while keeping a qualified human as the decision-maker.
The project adopts a theoretical model in which trauma (such as losing family, displacement, and direct violence) can lead to conditions such as PTSD, C-PTSD, anxiety, and depression, which in turn show up as observable non-verbal behavior: an avoidance pattern (expressionless face, low voice, avoiding eye contact) or a hyperarousal pattern (rapid movements, broken gaze, tense voice).
The model draws on indicators described in the literature, such as facial action units (FACS), gaze behavior, and vocal features like fundamental frequency, MFCC, jitter, shimmer, and HNR. This is the project's theoretical foundation; the actual analysis in the system is performed by the Gemini model, as described below.
The same model is used for all three tasks, with a different thinking level for each:
| Task | Model | Thinking level |
|---|---|---|
| Live streaming (real-time analysis) | gemini-3-flash-preview | Low, to reduce response time |
| Recorded video analysis | gemini-3-flash-preview | High, to improve analysis quality |
| Report generation | gemini-3-flash-preview | Medium, to balance speed and quality |
Other settings include a temperature of 1.0, a maximum of 8192 output tokens, high media resolution (media_resolution_high), and the application/json response type.
Nabd uses a multi-tier, client–server architecture:
| Screen | Function |
|---|---|
| Home page | Introduces the system and its services |
| Login and sign-up | Access by role (system administrator or therapist) |
| Dashboard | Statistics on children, sessions, and reports, with charts |
| Live analysis | Direct session with camera and microphone, live results and notes |
| Video analysis | Upload and analyze a recorded clip |
| Reports management | Search, filter, print, export, and track approval status |
| Child profiles | Basic data, status, sessions, and reports |
| Settings | Account, language, alerts, and security options |
The system administrator manages users, awareness articles, subscription plans, and system settings, while the therapist handles child profiles, sessions, analysis, and reports.
The system was tested in a simulated environment for live-assessment and video-analysis sessions, and the project documents the following operational results:
| Indicator | Documented result |
|---|---|
| Response time per analysis frame | 1.5 to 4 seconds |
| Capture rate in live analysis | One frame every 3 seconds (20 frames per minute) |
| Output format | Unified JSON converted into 0–100 indicators and Arabic labels |
| Simultaneous display zones | 7 interactive zones |
| Key performance indicators | 4: attention engagement, vocal activity, eye contact, social interaction |
| Network error handling | Retry up to 3 times at 2-second intervals |
The seven display zones cover the four indicators, the live emotional state, behavior analysis across four detail cards, charts for vocal activity and social interaction, real-time alerts, and a notes list updated both manually and automatically.
These are operational and technical results (speed, output consistency, and continuity of operation). The report does not compare the system's indicators against specialists' assessments or include a study with children in a clinical setting, so no diagnostic accuracy for detecting anxiety or PTSD is attributed to Nabd.
According to the outlook in the project, Nabd could be developed by:
Nabd processes video and audio recordings of children in fragile psychological situations, so any real use requires guardian consent, supervision by a qualified specialist, strict controls on access rights and data retention, and a review of the policy for sending media to cloud AI services. The system's results are supporting indicators, not a final medical diagnosis.
If you are working on an AI-driven behavioral-analysis or psychological assessment system and need technical help developing the idea or building a prototype, you can contact the Techno Enjaz office to discuss your project's requirements and the appropriate scope of support.
Nabd is a web system that supports the preliminary psychological assessment of war-affected children by analyzing non-verbal behavior and vocal cues with Google Gemini, developed as an academic project by a team of students with technical assistance from Techno Enjaz. It is a decision-support tool, not a substitute for clinical diagnosis.
The therapist starts a live session with camera and microphone or uploads a recorded video, and the Laravel server sends the data to Gemini, one frame every 3 seconds in live mode. Results come back as JSON and are converted into 0–100 indicators and Arabic behavioral labels, with real-time alerts and a draft report that the therapist reviews and approves.
The front end uses React 19 and TypeScript with Vite, Tailwind CSS, and Axios; the server is Laravel 12 on PHP 8.2; and the database is MySQL. The gemini-3-flash-preview model is used with a low thinking level for live streaming, high for video analysis, and medium for report generation.
In a simulated environment the system recorded 1.5 to 4 seconds per analysis frame with one frame captured every 3 seconds, and displayed 7 interactive zones including 4 key indicators (attention engagement, vocal activity, eye contact, social interaction). It retries up to 3 times on network failures, but the report includes no measurement of diagnostic accuracy against specialists.
The system was tested in a simulated environment rather than in clinics with real children, and there are no accuracy figures compared with specialists' diagnoses. It relies on the cloud-based Gemini service, so session media is sent to an external provider, meaning any real use requires guardian consent, specialist supervision, and strict privacy controls.
Reach out to Techno Enjaz engineering team to discuss technical feasibility, prototype hardware, and development scope.
Discover more engineering prototypes and academic implementations
A children's learning system that uses a camera to recognize animals with YOLO11, colors with HSV, and numbers with OCR, then shows and speaks the result in Arabic.
A computer vision system that detects knives and pistols in real time from a laptop or phone camera using two trained YOLOv8 models, Python, and OpenCV.
An interactive learning system for children that recognizes objects via the camera with YOLOv8, speaks their names, detects colors with HSV, and labels uploaded images.
