/ Changelog

What's new in ASKCECOS.

Every release, every feature, every fix — from first commit on 19 Feb 2026 to today. Built by CDGAI at CECOS University.

v1.0.012 May 2026

Feature

Public launch — ASKCECOS 1.0

Full admin panel live: conversation threads, smart handoff queue, knowledge base CRUD, dead letter queue, blocked numbers, analytics dashboard.
Auth surface overhaul: invite-only /request-access page, redesigned login with AuthShell primitive, NextAuth v5 JWT sessions (8h).
Landing site: homepage, About, Customers, Careers, Contact, Documentation, Security, Status, Changelog, Privacy, Terms, DPA.
WhatsApp AI pipeline in production: sub-2s response SLA, 12-step message processing, BullMQ job queue, dead letter replay.

v0.9.0May 2026

Improvement

KB retrieval overhaul — scope-gated, dominance-filtered

Added scope_tags column to knowledge base entries with GIN index; retriever now hard-filters by detected department (CS, Engineering, Business, Pharmacy, etc.) before ranking.
Dominance gate: if top result scores ≥ 1.5× the second, only the best match is sent to the AI — eliminates cross-department answer leakage.
Switched default model to gpt-4o-mini at reasoning_effort: minimal; fallback to gpt-4o only for borderline-confidence queries with KB context.
Prompt tightened: max_completion_tokens 512 → 220, temperature 1.0 → 0.3, history budget capped at 1,500 tokens per call.
Hybrid retrieval (pgvector cosine + lexical) shipped behind HYBRID_RETRIEVAL_ENABLED flag for future activation.
Frontend KB entry form: scope-tag chip selector and optional Urdu / Pashto Roman content fields.

v0.8.0Apr – May 2026

Feature

Admin panel — full feature set

Conversation threads with real-time SSE updates and message bubbles supporting text, images, audio, and documents.
Smart handoff queue: assign, tag, resolve, and track SLA per ticket.
Dead letter queue view: inspect failed WhatsApp deliveries and retry individually or in bulk.
Blocked numbers list: block by SHA-256 phone hash, unblock with audit trail.
Analytics dashboard: messages per day, auto-resolve rate, avg confidence, smart handoff breakdown — powered by Recharts via next/dynamic.
Settings panel: per-department confidence threshold tuning, pause/stop bot controls.

v0.7.0Mar – Apr 2026

Feature

12-step AI message pipeline

Full pipeline: dedup → block check → pause check → language detect → rate limit → attachment download → AI call → response parse → stall-phrase filter → confidence score → smart handoff decision → WhatsApp send.
Confidence-gated smart handoff: messages below threshold route to human agent queue instead of sending AI response.
Stall-phrase detector: 'ek minute', 'let me check', and similar filler phrases are discarded rather than sent.
Attachment handling: images, audio, and documents acknowledged inline (text-only AI pipeline) with MediaKit CDN storage.
BullMQ job queue with concurrency capped at 5 to protect OpenAI rate limits; DLQ for failed jobs.

v0.6.0Mar 2026

Feature

Knowledge base and multilingual support

Knowledge base CRUD: admins upload FAQs, policies, and academic calendars; entries are chunked and embedded for semantic retrieval.
Language detection: queries classified as English, Urdu Roman, or Pashto Roman; matched to appropriate reply template.
Static reply templates cover en, ur_roman, and ps_roman for all system messages.
Seven KB loaders: programs, fees, admissions, schedules, deadlines, FAQs, contact.

v0.5.0Mar 2026

Feature

WhatsApp Cloud API integration

Webhook endpoint: POST /webhook returns HTTP 200 in < 500ms, HMAC-SHA256 signature verification.
Outbound message sending via WhatsApp Business Cloud API v21.0.
Phone number privacy: raw numbers stored in Redis with TTL only; SHA-256 hash written to PostgreSQL.
Conversation session management with Redis-backed state and pause/stop-per-conversation controls.

v0.4.0Feb – Mar 2026

Feature

Database schema and backend foundation

PostgreSQL schema across 23 migrations: conversations, messages, tickets, knowledge_base, blocked_numbers, audit_log, staff.
node-pg-migrate with append-only, reversible migrations (up + down).
Redis 7 via ioredis: session state, rate-limit counters, real-time SSE fan-out.
Express 5 server with structured logging (no console.log), parameterised SQL only, Bearer token auth for admin API.

v0.1.019 Feb 2026

Feature

Project kickoff — monorepo scaffold

Monorepo initialised: backend/ (Node 20, Express 5, CommonJS) + frontend/ (Next.js 16, React 19, TypeScript strict, Tailwind v4).
CLAUDE.md autonomous operating manual, 9-agent workflow, docs/ and state/ directories established.
Core stack locked: OpenAI Chat Completions (text-only), BullMQ + Redis 7, PostgreSQL 16, WhatsApp Cloud API v21.0, ImageKit, NextAuth v5, TanStack Query v5.

/ Changelog

What's new in ASKCECOS.

Every release, every feature, every fix — from first commit on 19 Feb 2026 to today. Built by CDGAI at CECOS University.

v1.0.012 May 2026

Feature

Public launch — ASKCECOS 1.0

Full admin panel live: conversation threads, smart handoff queue, knowledge base CRUD, dead letter queue, blocked numbers, analytics dashboard.
Auth surface overhaul: invite-only /request-access page, redesigned login with AuthShell primitive, NextAuth v5 JWT sessions (8h).
Landing site: homepage, About, Customers, Careers, Contact, Documentation, Security, Status, Changelog, Privacy, Terms, DPA.
WhatsApp AI pipeline in production: sub-2s response SLA, 12-step message processing, BullMQ job queue, dead letter replay.

v0.9.0May 2026

Improvement

KB retrieval overhaul — scope-gated, dominance-filtered

Added scope_tags column to knowledge base entries with GIN index; retriever now hard-filters by detected department (CS, Engineering, Business, Pharmacy, etc.) before ranking.
Dominance gate: if top result scores ≥ 1.5× the second, only the best match is sent to the AI — eliminates cross-department answer leakage.
Switched default model to gpt-4o-mini at reasoning_effort: minimal; fallback to gpt-4o only for borderline-confidence queries with KB context.
Prompt tightened: max_completion_tokens 512 → 220, temperature 1.0 → 0.3, history budget capped at 1,500 tokens per call.
Hybrid retrieval (pgvector cosine + lexical) shipped behind HYBRID_RETRIEVAL_ENABLED flag for future activation.
Frontend KB entry form: scope-tag chip selector and optional Urdu / Pashto Roman content fields.

v0.8.0Apr – May 2026

Feature

Admin panel — full feature set

Conversation threads with real-time SSE updates and message bubbles supporting text, images, audio, and documents.
Smart handoff queue: assign, tag, resolve, and track SLA per ticket.
Dead letter queue view: inspect failed WhatsApp deliveries and retry individually or in bulk.
Blocked numbers list: block by SHA-256 phone hash, unblock with audit trail.
Analytics dashboard: messages per day, auto-resolve rate, avg confidence, smart handoff breakdown — powered by Recharts via next/dynamic.
Settings panel: per-department confidence threshold tuning, pause/stop bot controls.

v0.7.0Mar – Apr 2026

Feature

12-step AI message pipeline

Full pipeline: dedup → block check → pause check → language detect → rate limit → attachment download → AI call → response parse → stall-phrase filter → confidence score → smart handoff decision → WhatsApp send.
Confidence-gated smart handoff: messages below threshold route to human agent queue instead of sending AI response.
Stall-phrase detector: 'ek minute', 'let me check', and similar filler phrases are discarded rather than sent.
Attachment handling: images, audio, and documents acknowledged inline (text-only AI pipeline) with MediaKit CDN storage.
BullMQ job queue with concurrency capped at 5 to protect OpenAI rate limits; DLQ for failed jobs.

v0.6.0Mar 2026

Feature

Knowledge base and multilingual support

Knowledge base CRUD: admins upload FAQs, policies, and academic calendars; entries are chunked and embedded for semantic retrieval.
Language detection: queries classified as English, Urdu Roman, or Pashto Roman; matched to appropriate reply template.
Static reply templates cover en, ur_roman, and ps_roman for all system messages.
Seven KB loaders: programs, fees, admissions, schedules, deadlines, FAQs, contact.

v0.5.0Mar 2026

Feature

WhatsApp Cloud API integration

Webhook endpoint: POST /webhook returns HTTP 200 in < 500ms, HMAC-SHA256 signature verification.
Outbound message sending via WhatsApp Business Cloud API v21.0.
Phone number privacy: raw numbers stored in Redis with TTL only; SHA-256 hash written to PostgreSQL.
Conversation session management with Redis-backed state and pause/stop-per-conversation controls.

v0.4.0Feb – Mar 2026

Feature

Database schema and backend foundation

PostgreSQL schema across 23 migrations: conversations, messages, tickets, knowledge_base, blocked_numbers, audit_log, staff.
node-pg-migrate with append-only, reversible migrations (up + down).
Redis 7 via ioredis: session state, rate-limit counters, real-time SSE fan-out.
Express 5 server with structured logging (no console.log), parameterised SQL only, Bearer token auth for admin API.

v0.1.019 Feb 2026

Feature

Project kickoff — monorepo scaffold

Monorepo initialised: backend/ (Node 20, Express 5, CommonJS) + frontend/ (Next.js 16, React 19, TypeScript strict, Tailwind v4).
CLAUDE.md autonomous operating manual, 9-agent workflow, docs/ and state/ directories established.
Core stack locked: OpenAI Chat Completions (text-only), BullMQ + Redis 7, PostgreSQL 16, WhatsApp Cloud API v21.0, ImageKit, NextAuth v5, TanStack Query v5.