Prescreening Business Logic

Actors

Actor	Surface	Primary actions
Participant	dmind-prescreening-web-frontend	Consent, provide required details, choose an interview method, answer questions, and, when configured, view a result and select follow-up services.
Clinical staff	agnos-mental-dashboard-front	Authenticate, search and filter cases, review AI results, record 8Q/9Q assessments, add contact logs, close cases, and export CSV.
Corp/partner system	Prescreening backend integration routes	Create or validate sessions, pass participant details, receive progress, and handle redirects.
AI platform	AI controller and workers	Accept media/text answers and return depression, suicidal, two-question, and chief-complaint results.

Participant lifecycle

The exact path is feature-configuration and corp dependent. The frontend reads VITE_CORP_NAME and a JSON feature configuration; the backend reads its application name and Parameter Store feature configuration. The current participant frontend exposes video, audio, and text. CHOICE is deprecated and excluded from configured method selection, but its route and calculation code remain for compatibility. AIMET 9Q is deprecated and its page is not registered in the current route tree; the retained page submits the rating and then navigates to method selection, placing it before the interview in that implementation.

Participant API catalog

Phase	Endpoints	Maintained invariant
Entry	POST /interview/start, POST /interview/start/corp	Produces or reuses a session token and creates a case where required. Corp entry requires the corp ingress contract.
Session/content	GET /interview/validate-token, GET /interview/session, GET /interview/content	The Redis session determines the current/next question and resume state; content and question paths are cached from S3/configured assets.
Participant data	POST /interview/recipient, POST /interview/recipient-info	Stores the configured identity/demographic fields. Final contact info also shortens the session to the end TTL.
Assessment	POST /interview/submit-answer, POST /interview/submit-suicidal	Validates the configured question path and method, forwards answers to AI, and records the fallback source.
External and compatibility assessment	POST /interview/submit-self-rating, POST /interview/submit-multiple-answers, POST /interview/submit-self-rating-9q	The registered Mental Med/Agnos endpoint accepts external HAMD-7/8Q submissions as a pre-interview entry point; the current participant frontend does not call it, and external runtime use is unverified. Multiple-answer CHOICE and AIMET 9Q remain as deprecated compatibility paths.
Result	GET /interview/result	Returns AI, Agnos, fallback, and AIMET self-rating views. It persists a finished AI result and notifies a corp asynchronously when a corp_id exists.
Follow-up	GET /interview/terminal-service, POST /interview/select-terminal-service	Resolves currently available services, records the selection/time, and either returns a redirect or invokes the partner handoff.
1323	POST /interview/verify-alltime, POST /interview/validate-alltime-otp, POST /interview/resend-alltime-otp, POST /interview/alltime1323/progress, POST /interview/alltime1323/related-id	Maintains verification state in the session and progress/relationship data in the interview. Ingress progress/related-ID endpoints use the 1323-specific credential.
Surveys	POST /interview/feedback, POST /interview/low-depression-survey, POST /interview/sroi-survey	Stores user-experience, low-depression chatbot, and social-return data on the interview.
System	POST /system/refresh-content, POST /system/interview	Protected service-to-service maintenance/import entry points.

Session lifecycle

The participant session stores prediction_id, interview_id, corp_id, recipient name, current/next question, final-answer timestamp, default redirect, and 1323 phone/national-ID verification state. Default timings are:

• start TTL: 15 minutes
• active/default TTL: 120 minutes
• end TTL: 10 minutes

The interview document is durable; the Redis session is the resumable workflow cursor. A missing/expired session does not imply that the Mongo case was deleted.

Result source precedence

The stored depression_level is selected by source priority whenever a new source arrives:

Priority	Source	How it is produced
4	AGNOS_SELF_RATING	Partner-supplied Mental Med/Agnos HAMD-7/8Q result; external runtime use is unverified.
3	FALLBACK	Participant suicidal yes/no maps to SEVERE/LOW.
2	AI_SERVICE	Finished AI controller result.
1	AIMET_SELF_RATING	Deprecated, compatibility-retained local 9Q calculation.

A lower-priority result is still retained in its embedded field but does not replace the higher-priority top-level level/source. The AIMET 9Q thresholds are 0–6 normal/low, 7–12 mild/low, 13–18 moderate, and 19+ severe.

Corp behavior to preserve

The same source code serves multiple branded deployments. Corp-specific behavior includes required participant fields, invalid-token handling, result copy/actions, fallback URLs, available terminal services, analytics dimensions, and backend-to-partner credentials.

Current variants are main, TU/TU UAT, MDCU, CPIRD, THPF, KKU, Thangrath, and TSU. Navy is retained but inactive; Chula and Somdet remain legacy namespaces without a current service identified.

Terminal-service rules

Service	Eligibility/behavior in current code
ASA	TU only and requires a corp ID; backend starts an ASA flow and returns its redirect.
1323_DASHBOARD	TU, THPF, KKU, Thangrath, and TSU; copies the interview to main DMIND. TU/THPF/KKU require corp ID, while Thangrath/TSU use public entry.
1323_ALLTIME	Main, TU, THPF, KKU, Thangrath, and TSU; requires completed verification, persists the phone, and calls the AllTime validation/handoff.
HERE_TO_HEAL	Availability is resolved by its API using participant and AI context.
SATI, EMMIND, DMIND_CHATBOT	Availability/redirect is resolved by the configured client and stored in the interview's available-service snapshot.

TU additionally notifies its corp system of the chosen service. Feature configuration and frontend copy may narrow what a user sees, so preserve both backend eligibility and frontend configuration.

Staff dashboard lifecycle

The active dashboard is agnos-mental-dashboard-front; its production API base is https://api.prescreening.dmind.app/dashboard.

1. POST /auth/login returns access and refresh tokens for a MongoDB staff record.
2. The browser sends the bearer access token and uses POST /auth/refresh-token after unauthorized responses.
3. POST /interviews lists cases with status, province, severity, dates, exact case ID, and exact phone filters.
4. GET /interviews/:case_id returns participant, AI, staff-assessment, and contact-log detail.
5. POST /interviews/:case_id/result-staff stores the clinical follow-up result and can close a case.
6. Result-log endpoints create/edit/delete staff contact notes stored inside the interview record.
7. POST /csv exports the filtered result set.

Statuses are NEW, FOLLOW_UP, and DONE.

Dashboard query and case rules

• The list filters by status, province, severity, and created-date range. Request field record_id matches the exact numeric case_id; filter_text matches the exact phone number. When no phone filter is supplied, records with a missing, null, or empty phone number are excluded.
• Results sort by last_contact_date when present, otherwise created_at, with pagination applied after filtering/sorting.
• A contactable case requires non-negative staff 8Q and 9Q scores/severities. A not-accepted call or invalid phone clears those scores.
• Saving a staff result sets FOLLOW_UP; is_done or an invalid phone sets DONE. It updates last_contact_date and preserves the first staff-result creation time.
• Not-accepted calls increment not_accept_call_count and append an automatic Thai contact log. Free-form extra information also appends a log.
• Log edit increments is_edited; delete is a soft delete (is_deleted=true) by array index.
• Statistics group total/answered/not-answered counts by LOW, MODERATE, and SEVERE. CSV reuses the list filters without pagination.

Dashboard authentication

The backend validates active Mongo staff records and bcrypt-style stored passwords, issues access/refresh JWTs, and uses Redis refresh-token state for rotation/invalidation. Default access lifetime is 120 minutes and refresh lifetime is 24 hours. Registration is system-protected; case routes and auth-info/password-change require staff authentication.

Historical Agnos backend migration

The former Django backend stored mental records, medical records, and logs in PostgreSQL and linked them to DMIND through request/callback identifiers. Those dashboard responsibilities now live in dmind-prescreening-backend, with interview and staff state in MongoDB. Current-state changes must follow the Go/Mongo implementation; the old backend is reference material only.