Skip to content

UPES-ECS — Feature Roadmap (Next Horizon)

Scope: Features beyond what is already built or being built now, for the UPES Emergency Communication System — a LAN-only Asterisk PBX deployed on a van laptop (QEMU Ubuntu VM), with a web Operations Console.

Last updated: 2026-07-05

This document is deliberately forward-looking. It does not re-list the current backlog (which lives in Project-Status.md) — it catalogs the next wave of capability once the core is in production.


What already exists (do not re-plan)

  • Proven / built: 111 emergency hotline + ERT queue + escalation + voicemail + recording + incident logging; 199 drill; 198 echo; student-to-student SAP-ID calling; 700x paging (with access control); 9000x conference; responder pause/resume *45/*46; backup/restore; fail2ban; real TTS prompts; the Operations Console; a running QEMU PBX.
  • Being built now (out of scope for this doc): mass callout, roll-call / headcount, silent SOS / duress, pre-recorded announcements, on-call / follow-me routing, department hunt groups, dial-by-name directory, intercom, callback, and the Console v2 webapp.

  • Delivered 2026-07-05 (Console live-ops round):

  • Incident Timeline — live, newest-first stream of every handled call, built from the CDR log; emergencies / SOS / drills / bridges / paging colour-coded with filter chips. (Console section timeline, fed by status.json.cdr[].)
  • Presence & Shifts — registration + queue state of every defined endpoint (pjsip show endpoints), split into responder positions (on-shift / in-queue / reachable) and registered clients. (Console section presence, fed by status.json.presence[].) Supersedes the "Presence / BLF for responders" line below for the Console view (deskphone BLF still future).
  • Call Records + recording playback — recent call-detail log (roster-named) plus in-browser playback of whole-call recordings synced by Pull-Recordings.ps1. (Console section records, fed by status.json.cdr[] + recordings[].)
  • Insights / CDR analytics — emergency KPIs (answer-rate + answer-time on 111), drill pass-rate (199 answered/total over time), calls-by-type, activity-by-hour histogram, per-day volume, and top callers. Aggregated over the full Master.csv on the VM (python) → status.json.analytics{} → pure-CSS charts. (Console section insights, new Insights nav group.) Delivers the "CDR analytics dashboard" and "Drill pass-rate reporting" lines below.
  • 111 made answerable + offline panic-coach (102) — the ERT queue is now staffed by the on-shift registered handset (ert-shift.sh), so 111 rings a real phone (wallboard READY). And when nobody answers (queue timeout + escalation unanswered), the call now routes to a fully-offline, deterministic panic-coach (ctx_ai_helpline): a calm scripted decision-tree that coaches the caller through CPR, bleeding, choking, fire, lockdown, recovery position, and being trapped — with "9 = retry a responder" and "8 = leave a message". All audio is offline TTS (pico2wave+sox); no internet. Direct test dial 102. This is the zero-internet floor beneath the future AI triage (101, a local-first stack — Ollama/llama.cpp + Whisper (STT) + Piper (TTS), no Gemini/cloud — see ../AI-101/), and it satisfies SOP 19's rule that the fallback must work with every AI component offline.
  • Disaster-ready 111 flow (press-1 fast-path + parallel coaching) — grounded in Emergency Medical Dispatch practice (stay-on-line pre-arrival instructions, no redial loops). The old serial ring-out (queue → Lead 20s → backup 20s → then voicemail, ~45–60s of silence before any guidance) is replaced: the caller can press 1 any time during the queue to jump to first-aid (ctx_111_fastpath), and when no human answers the system does two things at once — rings the Lead + backup in the background (non-blocking call-files, alert_responders.sh, with "press 1 to join the queue") and starts the offline coach immediately + logs the incident. New prompts (real press-1 preanswer, periodic "press 1 for first-aid", responder alert). Full flow documented in ../Blueprint/03-Call-Flows.md §2, SOP 19 (de-vagued: 101/102 are internal routes, never a number a caller dials), and a new Emergency Call Flow Console section.

Everything below is the next horizon — sequenced, realistic, and mapped to real Asterisk capabilities.


How to read this

Every feature carries:

  • What it does — the capability in plain terms.
  • Why for a campus emergency system — the operational justification.
  • Asterisk / tech fit — the concrete mechanism (module, subsystem, or external component).
  • Effort — Quick (hours–days, config/scripting) · Medium (a week or two, new component or schema) · Project (multi-week, new hardware / subsystem / integration).
  • LAN-only? — yes (stays inside the closed campus LAN) · needs external (requires a controlled gateway, breaking the pure LAN-only posture — flagged explicitly).

LAN-only is a design invariant. Anything marked needs external must terminate on a single, hardened, one-way-where-possible gateway, never on the PBX itself. See Integrations.


1. Reporting & Analytics

Turns call data already flowing through Asterisk into evidence, trends, and after-action review.

Feature What it does Why for a campus emergency system Asterisk / tech fit Effort LAN-only?
CDR analytics dashboard Call-volume, per-extension, per-hour/day rollups; busy-hour and trend charts Shows load on 111, when incidents cluster, whether ERT is over/under-provisioned cdr_csv (already writing) or cdr_adaptive_odbc → SQLite/Postgres; charts in Console v2 Medium yes
Answer-time / SLA metrics Time-to-answer, ring duration, abandon rate on the 111 queue The single most important emergency KPI: how fast does a real call get a human? Queue events via CEL (Channel Event Logging) + AMI QueueSummary/AgentComplete; compute p50/p90 Medium yes
Missed-emergency trend board Aggregates the existing missed_incident.sh flags into a trended view Repeated misses are a safety failure that must be visible, not buried in a log file Read the incident logs / dashboard flags already produced by ../config/ scripts; render trend Quick yes
Drill pass-rate reporting Scores 199 drills over time: answered / escalated / missed, per shift Proves readiness to leadership; turns drills into a measurable program Tag drill CDRs (distinct context/accountcode) + join to incident IDs Quick yes
Recordings & incident review portal (restricted) Web portal to search, play, and annotate 111 recordings tied to their incident ID After-action review, training, and dispute resolution — with access tightly controlled Serve MixMonitor files behind RBAC; index by ERT-YYYYMMDD-NNNN from incident_id.sh; enforce SOP 13 retention Medium yes
Immutable audit log Append-only record of who listened to / exported / deleted recordings and who changed config Recordings are sensitive; access itself must be auditable. Config changes need a paper trail Structured audit table + hash-chain / append-only file; surface in Console; ties to SOP 26 Medium yes
Scheduled digest export Nightly/weekly PDF/CSV summary (volume, misses, drills) written to disk Gives IT/ERT leads a standing report without opening the Console Cron job over the CDR/CEL store; render locally (no cloud) Quick yes

2. Resilience & Continuity

Keeps the emergency line answering through power, hardware, and network failure.

Feature What it does Why for a campus emergency system Asterisk / tech fit Effort LAN-only?
Standby / failover PBX (van as failover) A second Asterisk that takes over if the primary dies — e.g. the van PBX backs up the fixed campus PBX (or vice-versa) An emergency line that goes down is the emergency. Redundancy is non-negotiable at go-live Two PBXes; failover via keepalived/VRRP floating IP, or phones with a backup registrar (aor outbound/secondary proxy); heartbeat health-check Project yes
Config auto-sync (campus ↔ van) Keeps dialplan, PJSIP endpoints, prompts, and roster identical on both nodes A failover node is only useful if its config isn't stale. Manual copy = drift = surprise on the worst day rsync/git push of /etc/asterisk + unixtimed reload; validate with asterisk -rx 'dialplan reload' dry-run; extends existing backup/restore Medium yes
Scheduled restore drills Automated, logged test that a backup actually restores and the PBX answers 198/199 after Backups that were never restored are hope, not a plan. Proves recoverability on a schedule Cron-driven restore into a scratch VM + automated 199 self-test; record pass/fail to the audit log; extends SOP 11 Medium yes
UPS / power monitoring Reads UPS state; alarms on mains loss and low battery; can trigger graceful shutdown Risk R1 — no power, no PBX. The van especially needs a monitored battery runway NUT (Network UPS Tools) on the host → alert into Console; optional dialplan alarm page via 700x; ties to SOP 21 R1 Medium yes
Watchdog & auto-recovery Detects a hung Asterisk / stuck registration and restarts the service Unattended van deployment can't rely on a human noticing a crash at 2am systemd restart policy + a liveness probe (SIP OPTIONS ping / AMI heartbeat) extending SOP 10 Quick yes
Cold-standby "PBX-in-a-box" image A pre-baked, one-command bootable image so a dead node is replaced in minutes Fast physical recovery when hardware (not just software) fails Golden QEMU image + Deploy-UpesEcsVm.ps1 (already exists) as the rebuild path Quick yes

3. Reach & Hardware — Audible Campus Alerting

Extends the system from phones-only to the whole campus can hear it.

Feature What it does Why for a campus emergency system Asterisk / tech fit Effort LAN-only?
IP speaker / PA integration Pushes live or pre-recorded audio to IP speakers / a PA head-end across campus People not holding a phone still need to be warned (evacuate, shelter, all-clear) SIP-endpoint speakers (Algo/CyberData-class) called as PJSIP endpoints, or RTP multicast paging (chan_multicast / one-way RTP) to a speaker group; hooks the "pre-recorded announcements" work being built now Project yes
Zone-based paging Address specific buildings/zones (block A, hostels, labs) instead of all-or-nothing Localizes response; avoids panicking the whole campus for a contained incident Per-zone paging groups (extension ranges / multicast groups) layered on the 700x paging plan (SOP 01) Medium yes
Multilingual announcements at scale (Hindi / English) Plays the correct-language prompt per zone, or bilingual back-to-back UPES is bilingual; an alert nobody understands is not an alert Pre-recorded prompt sets per language (extends SOP 28) selected by zone variable in the dialplan Quick yes
Multi-campus wireless bridge (Bidholi ↔ Kandoli) Extends the LAN across campuses over the rooftop point-to-point link One PBX can serve both campuses; ERT and paging reach Kandoli — already designed Layer-2/3 PtP bridge carrying SIP/RTP; failover-aware; already designed in SOP 20 (cross-link SOP) Project yes
Strobe / visual alert outputs Drives strobes/beacons alongside audio for high-noise or hearing-impaired areas Labs, workshops, and hostels at night need a visual channel too GPIO/relay via a networked I/O module triggered on page, or beacon-equipped IP speakers Medium yes

4. Provisioning & Scale

Moves onboarding from hand-config (today's error source) to automation that scales to a campus.

Feature What it does Why for a campus emergency system Asterisk / tech fit Effort LAN-only?
Bulk SAP-ID onboarding automation Ingests a roster CSV and generates PJSIP endpoints, secrets, and voicemail boxes en masse Thousands of students can't be added by hand; hand-config is the current top failure mode Template + generator over pjsip.conf/realtime; extends ../provisioning/pilot-users.csv flow; secrets set at import (never in repo) Medium yes
PJSIP Realtime (DB-backed accounts) Store endpoints/AORs/auth in a database instead of flat files Live add/remove without reloads; the scalable substrate for everything else in this section res_pjsip + res_config_odbc (PJSIP Realtime) → Postgres/MySQL on the PBX Project yes
QR self-provisioning kiosk Student scans a QR at a kiosk → phone auto-configures the correct, encrypted profile One scan replaces the transport/+91/encryption mistakes phones hit today Extends the QR/profile work in ../provisioning/linphone/; kiosk validates SAP-ID before issuing config Medium yes
MDM-managed ERT Androids Central management of the always-on ERT answer-point phones (app, config, lockdown, remote wipe) ERT phones must never sleep, must stay on the right app/config, and survive loss/theft MDM (e.g. Headwind/open-source) pushing the SIP client + SOP 24 battery/background hardening Project needs external*
Admin console for account lifecycle Create / suspend / role-change / offboard accounts with approvals, from the Console Turnover (students graduate, staff leave) needs a governed lifecycle, not ad-hoc edits Console v2 UI over PJSIP Realtime + AMI reloads; every change to the audit log; ties to SOP 04 role matrix Medium yes

* MDM typically wants an internet/cloud tenant. A self-hosted, LAN-only MDM keeps the invariant; a cloud MDM would need the controlled-gateway treatment.


5. Access & Security

Hardens signalling, media, roles, and secrets for a production emergency service.

Feature What it does Why for a campus emergency system Asterisk / tech fit Effort LAN-only?
TLS + SRTP (encrypted signalling & media) Encrypts SIP and RTP so calls can't be sniffed or spoofed on the LAN Emergency and duress calls are exactly the traffic you must not leak or tamper with (Risk H2/D3) res_pjsip TLS transport + media_encryption=sdes (or DTLS-SRTP); cert management; per-client profile decision — SOP 26 Project yes
Per-role RBAC hardening Enforces least-privilege on Console, portal, and dialplan features by role Not everyone should page campus, hear recordings, or change config. Blast-radius control Console RBAC + Asterisk contexts per role (already role-scoped in dialplan); align to SOP 04 / SOP 27 RACI Medium yes
SIP flood / DoS protection Rate-limits registration/INVITE storms and auto-bans abusers beyond today's fail2ban A flooded PBX can't answer 111. Availability is a security property here Tighten fail2ban (already present) + res_pjsip qualify/ACL + kernel nftables rate limits; AMI alarm on spikes Medium yes
Secrets manager Centralizes SIP passwords, cert keys, AMI/ARI creds out of flat files Secrets sprawl is a breach waiting to happen; rotation must be possible Local vault (e.g. pass/age/self-hosted Vault) feeding config at deploy; keeps repo placeholder-only (as today) Medium yes
Signed / verified config deploys Config only loads if it passes validation and a signature/integrity check Prevents a bad or malicious dialplan from silently breaking 111 Pre-reload dialplan reload dry-run + checksum gate in the deploy pipeline; audit every apply Quick yes

6. Operations & Responder Management

Day-to-day tooling so shifts, presence, and situational awareness run themselves.

Feature What it does Why for a campus emergency system Asterisk / tech fit Effort LAN-only?
Shift roster automation + auto-unpause Loads the ERT roster; auto-pauses off-shift responders and unpauses the on-shift ones at handover Removes reliance on responders remembering *45/*46; guarantees the queue always has live agents Scheduled AMI QueuePause/QueueUnpause driven by the roster; complements manual *45/*46; ties to SOP 30 Medium yes
On-call calendar Defines who is on-call by date/time and feeds routing + escalation Escalation must reach a real on-call person, not a fixed extension that may be off-duty Calendar → dialplan variables / follow-me targets (dovetails with the on-call routing being built now) Medium yes
Real-time control-room wallboard (TV) Full-screen live board on a control-room TV: active 111 calls, queue depth, longest wait, agent status, alarms Shared situational awareness during an incident; the operator shouldn't hunt for numbers Console v2 read-only wallboard view over AMI/CEL; auto-refresh; kiosk browser on the TV Medium yes
Presence / BLF for responders Shows each responder's live state (available / on-call / paused / offline) with busy-lamp fields Dispatch needs to see at a glance who can actually take the next call res_pjsip presence + dialplan hints → SUBSCRIBE/NOTIFY (BLF) on deskphones and in the Console Medium yes
Escalation-path visualizer & tester Shows and test-fires the 111 escalation chain without a real emergency Confidence that escalation will work before you need it; training aid Drive the existing escalation dialplan under a test context; log the path to the audit trail Quick yes

7. Integrations (BREAKS LAN-only — needs a controlled gateway)

Read first. Every item here crosses the campus LAN boundary. Each must terminate on a single, hardened, monitored gateway — never on the emergency PBX directly — with 111 staying fully functional even if the gateway is down. These are opt-in extensions, not core.

Feature What it does Why for a campus emergency system Asterisk / tech fit Effort LAN-only?
SMS / WhatsApp / email mass alerts Blasts an alert to students/staff over SMS, WhatsApp, and email in parallel with voice Reaches people who don't answer a call or aren't on a phone; multi-channel redundancy External notification gateway (SMS aggregator / WhatsApp Business API / SMTP relay) triggered by the PBX via AGI/ARI webhook Project needs external
PSTN dial-out to real 100 / 108 Bridges an on-campus incident to real police (100) / ambulance (108) Some emergencies must escalate to public services — the system should hand off cleanly SIP trunk to an ITSP or a DAHDI/analog FXO gateway; strict allow-list; caller-ID/CLI policy — note SOP 21 D1 100-collision risk Project needs external
CCTV / access-control pop-on-call On a 111 call, pops the nearest camera feed and unlocks/locks relevant doors for responders Cuts response time; gives the operator eyes on the scene instantly ARI/AGI event → VMS API (ONVIF) + access-control API on the gateway LAN segment Project needs external
SIS / HR directory sync Auto-populates the dial-by-name directory and roster from campus SIS/HR Keeps the directory correct as people join/leave — no manual maintenance Scheduled one-way pull from SIS/HR API → PJSIP Realtime; feeds the dial-by-name work being built now Project needs external
Native UPES VoIP app (GPS / coordinate broadcast) A branded app that registers over SIP, broadcasts caller GPS/coordinates on an emergency, and rings even backgrounded via push Directly answers Risk R5 (location blindness); push fixes the "phone asleep, call missed" failure Custom SIP client (PJSIP mobile SDK) + push (FCM/APNs) waking it on INVITE; coordinates ride in headers/AGI into the incident record Project needs external
App-push wake-on-INVITE Wakes a backgrounded/sleeping phone so an incoming emergency call actually rings Battery-optimized Androids silently drop SIP; without push, calls to sleeping phones are lost (SOP 24) res_pjsip push params (PJSIP-endpoint push) + a push gateway (FCM/APNs) — the wake path is inherently cloud Project needs external

8. Accessibility & Inclusion

Ensures people who can't make a normal voice call still reach help.

Feature What it does Why for a campus emergency system Asterisk / tech fit Effort LAN-only?
Text / SMS-to-emergency Lets someone report silently by text when they can't or shouldn't speak Duress, hearing/speech disability, or a situation where talking is dangerous LAN messaging (SIP MESSAGE / in-app chat) into the ERT workflow; SMS variant needs the gateway (§7) Medium yes (SMS: needs external)
TTY / RTT support Carries real-time text / teletype so deaf/HoH users converse in an emergency Accessibility obligation; some users have no voice option at all Asterisk RTT (T.140) over the SIP session where clients support it; document supported endpoints Project yes
Louder / vibrate answer-point handling High-volume ring + strong vibrate + visual flash on ERT answer points ERT phones in noisy/pocketed conditions must never be missed Client-side profile on the always-on ERT Androids (via MDM, §4) + SOP 24 hardening Quick yes
Accessible Console (WCAG) Screen-reader-friendly, high-contrast, keyboard-navigable Operations Console Operators and admins may themselves have accessibility needs Console v2 front-end work: semantic HTML, ARIA, contrast, focus order Medium yes

A pragmatic shortlist for the first push after the current backlog and the being-built features land. Chosen for safety impact per unit of effort, and for staying LAN-only.

  1. CDR + answer-time analytics dashboard (§1, Medium, LAN-only) — You already generate the data; you're flying blind without the view. Answer-time on 111 is the emergency KPI, and drill/missed-incident trends turn readiness into something you can show leadership. Highest insight-to-effort ratio.
  2. Shift roster automation + auto-unpause (§6, Medium, LAN-only) — The queue is only as good as the agents logged into it. Automating pause/unpause at handover removes a human single-point-of-failure (someone forgetting *45/*46) that can silently gut 111.
  3. UPS / power monitoring (§2, Medium, LAN-only) — Directly retires Risk R1. On a van, an unmonitored battery is a countdown to a dead emergency line. Monitoring + graceful shutdown is modest effort for outsized reliability.
  4. IP speaker / zone paging (start small) (§3, Project but stageable, LAN-only) — The biggest reach gap: today only phone-holders get alerted. Even one pilot zone of IP speakers proves audible campus-wide warning and evacuation — the capability people intuitively expect from an emergency system.
  5. TLS + SRTP (§5, Project, LAN-only) — Before any wider go-live, emergency and duress traffic must be encrypted. It closes Risk H2/D3 and is a prerequisite for trusting the system with real incidents. Bigger lift, but non-negotiable pre-scale.

Deliberately deferred from the top 5: anything in §7 (SMS/WhatsApp, PSTN 100/108, native app, push) — high value but each breaks LAN-only and needs the controlled-gateway decision first. Tackle them as a governed second phase, not opportunistically.


Note on AI-101

AI-101 (local-first — Ollama/llama.cpp + Whisper + Piper, no cloud/Gemini) — an AI triage front end on extension 101 — is fully designed in ../AI-101/ and SOP 19, but is deferred by decision. 111 stays human-first and must keep working even if every AI component is offline. It is intentionally not on this roadmap's active track.