Home / AI Helpdesk

Human-Taught AI Helpdesk

Egypt Hub - Curated Knowledge Base and RAG System

Objective

Deploy a human-taught, non-self-learning AI Helpdesk, managed and curated from Egypt, to support all Odoo 19 users (internal teams, students, and clients).

This system blends LangChain-based RAG (Retrieval-Augmented Generation) with Odoo's Knowledge and Helpdesk modules—ensuring transparency, accuracy, and human control.

Quick Launch Checklist

1

Egypt Hub Setup

Stand up Egypt AI Helpdesk Hub: Odoo Helpdesk + LangChain + Vector DB (Postgres/FAISS/Chroma)

2

Human-in-the-Loop

Implement human-in-the-loop curation loop: only approved Knowledge articles feed the AI index

3

LangChain Middleware

Build LangChain middleware microservice for retrieval, logging, and human fallback

4

RACI & Governance

Design RACI & governance structure (Curators / Engineers / Auditors)

5

Auditability

Maintain auditability: every AI answer cites KB Article ID + Version

6

Operational Runbook

Establish runbook: daily/weekly ops, content updates, escalation, and SLA monitoring

A. System Architecture (Schematic Outline)

                 ┌───────────────────────────────┐
                 │        Odoo 19 Instance        │
                 │  Helpdesk + Knowledge + Docs   │
                 └──────────────┬────────────────┘
                                │  (API / ORM)
                                ▼
┌──────────────────────────────────────────────────────────┐
│         LangChain Middleware (Python FastAPI Service)     │
│-----------------------------------------------------------│
│ • Ingest Approved KB articles (nightly)                  │
│ • Create vector embeddings (OpenAI / Mistral / Gemini)   │
│ • Store in VectorDB (FAISS / Chroma / Postgres pgvector) │
│ • Query pipeline: RAG → Re-rank → Response generator     │
│ • Log: question, source IDs, curator, confidence score    │
└──────────────────────┬────────────────────────────────────┘
                       │
         ┌─────────────┴─────────────┐
         │ Human-Taught Review Loop  │
         │ (Egypt Curator Console)   │
         └─────────────┬─────────────┘
                       │
             ┌─────────┴─────────┐
             │    User Front-End │
             │  (Odoo Portal UI) │
             └───────────────────┘

Key Tech Stack

LangChain + Python FastAPI for RAG logic FAISS / pgvector for embeddings storage Odoo REST / RPC API for KB synchronization Audit Logs DB (PostgreSQL) for traceability

B. Process Flow (Stepwise)

1

Article Creation

Expert writes KB article in Odoo ("Draft")

→
2

Peer Review

Second expert reviews; status → "Reviewed"

→
3

Approval

Knowledge Lead approves; status → "Published"

→
4

Nightly Index Job

LangChain middleware ingests all Published articles and rebuilds the vector store

User Query Flow

  1. User Query (via Helpdesk or Portal): AI fetches top-3 relevant articles → generates summarized response + citations
  2. Response Displayed in chat widget or ticket view
  3. Human Feedback: If "Not helpful," ticket reopens to human curator
  4. Curator Updates: Curator updates article or creates new one; re-approval triggers re-indexing
  5. Audit Trail: Every AI response logged: {ticket_id, article_id, version, confidence, responder_type}
  6. Reports Reviewed weekly by Helpdesk Manager + Audit

C. Governance Table

Role Responsibility Toolset / Access KPIs
AI Helpdesk Lead (Egypt) Owns helpdesk operations & curation cycle Odoo Helpdesk + LangChain Console SLA adherence ≥ 95%, AI coverage ≥ 80%
Human Curator Review / approve KB content; resolve escalated tickets Odoo Knowledge + Docs Avg. review time ≤ 2 days
LangChain Engineer Maintain middleware, monitor logs, fix pipelines FastAPI / LangChain server Uptime ≥ 99%, error rate < 2%
Internal Audit Monthly validation of log integrity Audit Dashboard Zero missing citations
IT Ops Infrastructure & backups Odoo.sh / K8s RPO 1 hr / RTO 4 hr
University Partner Coordinator Upload learning content & FAQs eLearning + Knowledge ≥ 50 new curated articles / semester

D. Content Lifecycle & Quality Gates

Stage Status Field Gate Criteria Action / Owner
Draft draft Article written; awaiting peer review Author
Review in_review Reviewer assigned; feedback logged Peer Reviewer
Approved approved Verified for accuracy; tagged by topic Curator Lead
Published published Live in Odoo KB; eligible for AI index AI Lead
Archived archived Superseded; excluded from index Curator Lead

Automatic Triggers

E. Human-in-the-Loop Controls

Control ID Description Mechanism (Odoo + LangChain) Owner
HIL-1 AI cannot train autonomously Disable learning; index static KB only AI Engineer
HIL-2 Every AI answer cites source Append [Article ID – Version] in response footer LangChain middleware
HIL-3 Confidence threshold < 0.70 → Escalate to human curator Middleware rule
HIL-4 Sensitive data screening Regex / PII detector pre-output Middleware
HIL-5 Manual override Curator can lock article from AI index Odoo flag exclude_ai=True
HIL-6 Feedback loop "Helpful?" button updates feedback score Helpdesk UI
HIL-7 Weekly QA audit Sample 20 AI responses / week AI Lead + Audit

F. Runbook – Operational Routine

Frequency Task Owner Tool
Daily Review new tickets & AI responses Curators Odoo Helpdesk
Re-index LangChain KB (02:00 Cairo) LangChain Engineer FastAPI Job
Weekly Quality audit of AI answers AI Lead + Audit Audit Report
Backup vector DB + logs IT Ops Odoo.sh / pg_dump
Monthly Curator training session HR / Edu Coord. eLearning
SLA & AI coverage report to CFO AI Lead Dashboard
Quarterly Security review + penetration test IT Ops / Audit Checklist
Annually AI ethics & bias review Governance Board Report

G. Integration Points (Technical Endpoints)

System Direction Interface / Method Description
Odoo → LangChain Outbound REST GET /api/articles?status=published Fetch approved articles for indexing
LangChain → Odoo Inbound REST POST /api/tickets/{id}/response Push AI answer with citations
Helpdesk → LangChain Query WebSocket / REST POST /query User question payload
LangChain → Audit DB Log SQL INSERT / API Store metadata (ticket, article, confidence)
BI Dashboard → Odoo Pull Odoo Spreadsheets / JSON API Display KPI metrics

H. Security & Compliance Controls

Area Control Implementation
Data Privacy No personal/financial data in index Filter fields before ingestion
Access Control RBAC for Curators, Engineers Odoo groups + VPN/IP whitelist
Traceability Immutable log of AI responses Audit DB + SHA-256 hash
Confidentiality TLS 1.3 / AES-256 at rest Standard in all data paths
Disaster Recovery Vector DB + Logs backup every 12 h DR Plan (Section 2 L)

I. Metrics & Dashboards

KPI Formula / Source Target Owner
AI Coverage Rate Auto-answered / Total Tickets ≥ 80% AI Lead
Human Escalation Rate Escalated / Total Tickets ≤ 20% Curators
Response Accuracy Score % "Helpful" feedback ≥ 90% Curators
Article Freshness Index # articles < 90 days old / total ≥ 70% Curators
SLA Compliance Resolved ≤ 2 business days ≥ 95% AI Lead
Downtime Unavailable minutes / total minutes ≤ 0.5% IT Ops

J. Error Handling & Escalation

Error Type Auto Detection Action Escalation To
API Timeout / 500 LangChain logs Retry × 3, alert Slack #ai-alerts Engineer
Invalid Article Ref Middleware validation Skip + flag in log Curator Lead
No Results (confidence < 0.7) Query result Escalate to human curator ticket Curator
PII Detected Regex filter Block response + alert Compliance Audit
Index Job Failed Cron log Rollback + retry + alert Engineer
SLA Breach Dashboard Auto-email CFO + AI Lead Governance Board

K. AI Helpdesk Audit Package (Monthly)

Contents archived under /Audit/AI_Helpdesk/<YYYY-MM>/:

  • Response Log Export (CSV)
  • Curator Approval Records (PDF)
  • System Health Report (JSON)
  • SLA & KPI Dashboard Screenshot
  • Exception Tickets Summary (Helpdesk CSV)
  • Internal Audit Sign-Off

Retention: 7 years

✅ Validation Check – Format & Structure
  • Architecture schematic + process flow included
  • Governance & RACI table provided
  • Runbook (daily-annual tasks) detailed
  • Integration, security, metrics, and error tables complete
  • Fully compliant with blueprint artifact structure; deployment-ready
← Previous: Education Next: Organization →