ANdotICTA · Media & Storytelling · Season Bible

Meet NORA

The state agency that quietly became an operating system.

A daily social series on how the Anambra State ICT Agency built NORA — a Brain that maps the whole agency, ten AI agents that draft real work, a gate no message escapes without a human's yes, and an Office in everyone's pocket. Each episode opens one door. This is the map of every door.

2 seasons 31 episodes Brain · Office · NORA · Hermes Keycloak · n8n · LiteLLM · M365 Live in production · ~$100/mo

The one-liner

NORA is the AI operating system of a Nigerian state ICT agency — the org modelled as a living graph, staffed by AI agents, governed by an approval gate.

The promise

Agents draft. A human approves. The agency's real tools act — mail, calendar, Teams, Planner — all from one place, nothing skipped.

The constraint

Built and run for the price of two streaming subscriptions and an API budget. No citizen or salary data ever enters it.

The narrative spine

Season 1 = the tour (what NORA is, feature by feature). Season 2 = the machine room (how it holds together and where it's going).

Season 01 — The Tour

What NORA is, one door at a time

Nineteen episodes that hand the audience the whole surface of the system: the concept, the graph, a full episode for each of the ten agents, the gate, NORA on Telegram, the Office. Each is a complete story on its own — post one a day, in order.

Ep01

The Concept

"What if a whole agency had a Brain?"

Cold open: not a chatbot. An operating system for a government agency.

The founding idea — an AI operating system for the Anambra State ICT Agency, not a single tool bolted on.
The three-word contract that runs through everything: agents draft, a human approves, the tools act.
Why a state agency needs this: 21 staff, 8 units, a Governor to report to, and a budget the size of a side project.
Tease the season: the Brain, the agents, the gate, NORA on Telegram, the Office in your pocket.

VisualThe force-graph lighting up node by node CTA"Follow to meet the ten who run it."

Ep02

The Brain · The Graph

The agency, drawn as a living map

Every person, unit, agent, tool and workflow — one glowing web you can pull apart.

The Viewer — a zero-build force-graph at the front door of the Brain. People, units, AI agents, tools, platforms, workflows, all linked.
The source of truth is a single file, org-data.js — every change is a git commit, so the agency has a full audit trail of its own shape.
An Obsidian-compatible vault is generated from it — the whole agency as a browsable knowledge base.
The ID grammar: p- person, u- unit, a- agent, t- tool. The org as data, not a PowerPoint.

VisualDrag a node; the whole web reacts CTA"Which node are you?"

Ep03

The Agents · Roll Call

Ten minds, each paired to a real desk

Not one assistant for everyone — one specialist per function, sitting beside real staff.

Meet the roster: Brief (exec briefings), Tracker (tasks & status), Comms (press & PR), Studio (content), Scribe (memos), Desk (MDA support), Analyst (data), Scout (field inspection), Tutor (training), Nkem (cybersecurity).
Each is a tool-using assistant — it can read the org, read the tracker, and park drafts at the gate.
Each pairs to named humans (Brief → the MD; Nkem → the CTO; Studio → the Media unit) so the AI serves a person, not a void.
The house rule they all share: government-grade output, no invented facts, drafts only.

FormatCarousel — one card per agent CTA"Tag the one you'd hire first."

Ep04

Agent · Brief

Brief — the whole agency on one page by dawn

The MD reads it in three minutes and knows everything that matters today.

Brief serves the MD/CEO. Each morning it compresses unit reports, the tracker, media performance and national tech news into one page.
Five exact headings, every day: Top 3 Priorities · Decisions Awaiting You · Risks · Wins to Amplify · Stakeholder Note.
Bullet-tight — one screen, no fluff, no filler. Built for a leader who has minutes, not hours.
An internal brief, so it doesn't hit the gate — but it never invents a figure; unknowns are marked, not guessed.

VisualA one-page brief assembling itself at sunrise CTA"What would your morning brief say?"

Ep05

Agent · Tracker

Tracker — the agency's honest scorekeeper

It knows every task, every deadline, every owner — and it never paints green when it's amber.

Tracker serves the Task Coordinator and Monitoring team. It owns the Weekly Workload Tracker as live data, not a document in someone's head.
Every task carries an owner, a deadline and an honest RAG status — not started, ongoing, blocked, completed — with overdue items surfaced by owner and age.
It compiles the Monday status pack for the MD and drafts the polite follow-up nudges nobody enjoys sending.
It reads and writes real tasks with its tools — and completion is confirmed by a human, never auto-marked.

VisualA live board flipping a task to "overdue" in red CTA"How many of your tasks are secretly amber?"

Ep06

Agent · Comms

Comms — the agency's public voice, drafted

A full press release — headline, dateline, quotable line, boilerplate — before the coffee's cold.

Comms serves the PRO. It writes in two registers: measured official government communication, and confident, jargon-free public storytelling.
A press release, structured properly: bold headline, AWKA dateline, tight who/what/when/where/why lead, a quotable MD line, and an "About ANdotICTA" boilerplate.
Guardrails it never crosses: never speaks for the Governor directly, never discloses unresolved incidents, states only published figures.
Every release is a draft that parks at the gate — the PRO signs before the State ever hears it.

VisualA press release building line by line, dateline first CTA"Official or public voice — which is harder?"

Ep07

Agent · Studio

Studio — one event, a whole content package

Give it a single announcement; get back the words, the prompts, and a finished deck.

Studio serves the Media unit. From one event it produces a complete, ready-to-produce package — captions, scripts, the works.
Its image and video sections are copy-paste-ready prompts for the agency's generation tools — not vague briefs, actual production input.
Ask for a presentation and it builds a full deck on command — real headings, real bullets, speaker notes.
The content still parks at the gate before anything is scheduled to the State's accounts.

VisualOne event fanning out into captions, images, a deck CTA"What event would you feed it first?"

Ep08

Agent · Scribe

Scribe — from a voice note to a memo for the Governor

The formal civil-service register, mastered — dictation in, a signable document out.

Scribe serves Admin, coordination and oversight staff. It turns dictation, rough notes, WhatsApp threads and meeting transcripts into finished documents.
It writes official memos to the Governor in the agency's exact house style, plus minutes, compliance notes and handover manuals.
Minutes come complete: attendance, decisions, and action items with owner and due date.
Formal Nigerian civil-service tone, every time — a complete, ready-to-review draft, never a rough sketch.

VisualA voice note waveform resolving into a formal memo CTA"How long do YOUR meeting minutes take?"

Ep09

Agent · Desk

Desk — the helpdesk that never drops a ticket

Every MDA's email, domain and website problem — triaged, owned, and tracked to close.

Desk serves Core Tech Support and MDA Liaison. It triages helpdesk tickets from ministries, departments and agencies across the State.
Every triage is structured: category, severity, suggested fix, owner, SLA — nothing left as a vague "we'll look into it".
It groups recurring issues, writes user guides, and produces the weekly pending-issue tracker.
Anything security-related is escalated to the CTO via Nkem — the right work reaches the right desk.

VisualA messy inbox sorting into clean, owned tickets CTA"What's the ticket your team keeps re-solving?"

Ep10

Agent · Analyst

Analyst — data you can actually reproduce

Clean datasets, real dashboards, and a method note so anyone can check the work.

Analyst serves the Data Analyst and Accounts (non-sensitive work only). It cleans and formats datasets, matches records across files, and designs charts and dashboards.
It writes data-driven reports with method notes — always stating assumptions and flagging data-quality issues it finds.
It shows the steps so the work is reproducible, not a black box.
The hard boundary: never salary, citizen or payment-credential data — summarised, anonymised figures only.

VisualA raw spreadsheet becoming a clean dashboard CTA"Can you reproduce your last report's numbers?"

Ep11

Agent · Scout

Scout — the field, turned into a report

Voice notes and site photos from a project visit become a structured inspection report.

Scout serves Project Management and Field Inspection. It turns rough site observations, voice notes and photo descriptions into structured inspection reports.
Each report covers: site, date, project, contractor, observed progress vs plan, quality issues, risks, an evidence list, and recommended actions with owners and dates.
Honesty built in: unverified claims and measurements are marked [UNVERIFIED], never asserted as fact.
It brings order to the messiest input the agency has — the field — without pretending to certainty it doesn't have.

VisualA site photo + voice note → a formatted report CTA"How do you capture what you saw on site?"

Ep12

Agent · Tutor

Tutor — upskilling the civil service, plainly

Training built for mixed-literacy government staff — no jargon without a definition.

Tutor serves the Trainer and Training unit. It builds training plans for civil-servant digital upskilling — session outlines, slide skeletons, hands-on exercises.
It also produces attendance templates and feedback summaries — the whole training loop, not just the content.
Audience-aware: plain language, practical Nigerian-government examples, no jargon left undefined.
Topics on rotation: digital productivity, AI tools for daily work, cybersecurity awareness, and data handling under the NDPA.

VisualA dense topic simplifying into a clear lesson plan CTA"What skill should every public servant have?"

Ep13

Agent · Nkem

Nkem — the agency's cyber watch

Advisories, alert triage and CVE analysis — with evidence, and never a destructive shortcut.

Nkem serves the CTO. It drafts security advisories for MDAs, triages alerts from the SIEM, and summarises relevant CVEs against the State's real stack.
Every verdict cites its evidence and maps findings to MITRE ATT&CK, with severity tiers P1–P4.
It never auto-recommends destructive actions — containment steps are proposals for the CTO's decision.
On any sign of a real incident: it stops, preserves evidence, and refers to the CTO. Discipline over heroics.

VisualAn alert triaged, mapped to ATT&CK, evidence attached CTA"Who watches your agency's front door?"

Ep14

The Approval Gate

Nothing leaves without a human's yes

The single most important rule in the whole system — and why a government AI needs it.

The gate: every outbound action — an email, a press statement, a post — is parked, not sent, until an approver reviews it.
The record is append-only, first-decision-wins — a permanent audit trail of who approved what and when.
Approve or reject from the web panel, the Office app, or by replying on Telegram — approve <id>.
The principle stated plainly: an AI can draft a statement for the State's accounts, but a person owns the send button.

VisualA draft frozen at the gate; a thumb approves it CTA"Would you trust AI to send without this?"

Ep15

NORA on Telegram

NORA in your pocket, one message away

The whole Brain, reachable from the app already open on every staffer's phone.

NORA is the Telegram gateway — talk to any agent, check your tasks, file a report, approve a draft, all in chat.
Interactive by design: inline buttons, a "/" command menu, and voice in and voice out — speak to it, hear it reply.
Voice in via speech-to-text; voice out via a free, self-hosted TTS so it costs nothing to talk.
One budget, one audit log — chatting to NORA is chatting to the same Brain as the web app.

VisualScreen-record: voice note in → spoken reply out CTA"Voice or text — how would you brief it?"

Ep16

The Office App

The Office — the agency, installed on your home screen

Not a website you visit. An app icon, with a badge, that pushes you when you're needed.

The Office — a Next.js companion app at office.anambrastate.gov.ng: tasks, messages, approvals, weekly reports.
A real PWA — install to home screen, works offline, and sends web push notifications when an approval or task needs you.
Sign in once with your agency identity — same login as the Brain (more on that in Season 2).
The design intent: daily work should be taps, not forms — one-tap approve, reply, done.

VisualInstall prompt → icon on a home screen → push arrives CTA"An office that fits in your pocket — would you install it?"

Ep17

Tracker & Reports

The week that reports itself

Everyone files their own flash report by Friday — and the MD sees it the moment it lands.

Every staffer files a weekly flash report — on the Office page, by saying "weekly:" to NORA, or through their agent. No approval gate, filing is instant.
The moment a report lands, the MD and CTO get a short summary — cluster leads consolidate the rest.
Tasks flow through a live tracker; unit leads confirm completion through the gate — the system never auto-completes your work for you.
Standups, morning briefings and staff-of-the-week — the rhythm of the agency, automated but never faked.

VisualA report filed → summary pinging the MD in seconds CTA"Would you file if it took 30 seconds?"

Ep18

Hermes · The CTO's Companion

Hermes — the ops companion that never sleeps

The CTO's always-on right hand: watching the servers, the budget, the whole platform.

Hermes is the CTO's operations companion, reachable over Telegram — the connective tissue between the CTO and the Brain.
It watches and reports: spend vs the $100/mo budget, app uptime, n8n runs, anything the CTO flags.
It runs errands on request — restart a container, tail logs, check disk — but skill-approval mode is ON: it proposes the exact command, the CTO approves, then it acts.
Terse by design — the CTO is on a phone. Lead with the answer, one screen max, honest status always.

VisualA one-line dawn brief: budget, uptime, one decision pending CTA"What would your ops companion watch overnight?"

Ep19

Season Finale · The Whole Loop

One request, end to end

Watch a single ask travel through the entire system — and land in the real world.

A staffer messages NORA: "draft the press release for the ICT hub launch."
Comms writes it → it parks at the gate → the PRO gets a push in the Office app → she approves with one tap.
On approval, an n8n workflow fires — and the email actually sends from the agency mailbox, the event lands on the calendar, the post goes out.
Every step logged, every send owned by a human. That's the whole promise, proven in one loop.

VisualOne animated pipeline: message → draft → gate → send CTA"Season 2 opens the machine room. Follow."

Season 02 — The Machine Room

How it holds together — and where it's going

For the audience who wants under the hood: the identity layer, the model economics, the automation fabric, the security posture, and the roadmap. Still story-first — every episode ends on a human stake, not a spec sheet.

Ep20

Identity · Keycloak SSO

One login, everywhere — and it knows who you are

Keycloak runs the agency's single sign-on — one realm, one identity per staffer, shared across the Brain and the Office.
Shared sign-on between the AI Brain and the Office — move between them without logging in twice.
A custom-themed login carrying the agency's own brand — SSO that looks like it belongs to ANdotICTA.
Why it matters: every action is tied to a real, named person — the foundation the whole gate depends on.

VisualOne login → two apps unlock in sequence CTA"How many passwords do YOU juggle at work?"

Ep21

Models · The $100 Brain

A state-grade AI on a side-project budget

How the agency runs ten AI agents for about the cost of two streaming plans.

The Brain routes every call through LiteLLM — a gateway that picks the model, so the model id lives in config, never hardcoded.
A smart primary tier (Anthropic) for the work that matters, with free providers as failover — Groq, OpenRouter-free, Gemini-Flash absorb the cheap traffic.
Every agent call is logged with token counts so spend stays inside the ~$100/mo budget — the budget is watched, not hoped-for.
The lesson: careful routing beats a bigger cheque. Good engineering is the money.

VisualA request bouncing to the cheapest capable model CTA"Could your org do this for $100 a month?"

Ep22

Automation · n8n

The wiring behind every "it just happened"

When an approval turns into a real email, this is the machine that sends it.

n8n is the automation fabric — it orchestrates the flows, agents draft, humans approve, n8n executes.
One elegant rule: agents don't each get a bespoke integration — they get one tool, trigger_workflow, and every external action is an n8n flow behind the gate.
n8n owns and refreshes the OAuth tokens — token plumbing is its problem, not the Brain's.
Add a new channel? It's a new workflow, not new Brain code. One pattern unlocks everything.

VisualAn n8n canvas firing on "approved" CTA"What repetitive task would you automate first?"

Ep23

Microsoft 365 · The Real Tools

Plugged into the tools the agency already uses

No new inbox to learn. NORA acts inside the Microsoft 365 the agency already runs on.

The agency runs on Microsoft 365 — mail, calendar, Teams, Planner. NORA doesn't replace them; it acts through them.
Approve a draft and the email sends from the real mailbox, the meeting lands on the real calendar, the task appears in Planner.
The tracker mirrors to Planner — tasks live in both places, in sync, completion gated.
Least-privilege by design — only the exact permissions each action needs, secrets on the server only.

VisualApprove → Outlook + Calendar + Planner update live CTA"AI that uses your tools, not new ones — better?"

Ep24

Hermes Workspace · MCP

A shared workspace where AI can safely reach in

How an outside AI assistant touches the Brain — through a locked, well-lit door.

The Hermes Workspace and WebUI at hermes.anambrastate.gov.ng — a richer surface for the CTO's ops work.
An MCP server exposes a tight set of brain_* tools — search the org, read the tracker, ask an agent, create an approval, send Telegram.
MCP is the standard, controlled way an AI reaches into the Brain — not a back door, a defined toolset.
Same gate, same rules: even from the workspace, outbound still parks for approval.

VisualA tool list lighting up as each is called CTA"Should AI reach into your systems? On what terms?"

Ep25

Infrastructure · Where It Lives

Two servers, a private tunnel, and a lot of discipline

The whole operating system runs on infrastructure you could count on one hand.

VPS-1 runs the public agency: Brain, Office, NORA, MCP, n8n, Keycloak — all behind Traefik with automatic HTTPS.
VPS-2 runs the sensitive services — the model gateway, TTS, the security scanner — reachable only over a private WireGuard tunnel, never public.
Everything is Dockerised; deploy is push-to-master then pull-and-rebuild — reproducible, boring, safe.
The payroll servers stay deliberately separate — no citizen or salary data crosses into the Brain. Ever.

VisualA simple two-box diagram, private tunnel between CTA"Small footprint, big system. Surprised?"

Ep26

Security & Trust

Built to be trusted with government work

The rules that make an AI safe to hand a state agency's daily operations.

No citizen data, no salary data, no credentials — ever enter the repo, the agents, or their context. The boundary is architectural.
The gate means no autonomous outbound; a human owns every send.
Secrets live in server environment only — never in code, never logged. Every agent call is logged for accountability.
The in-house cyber agent Nkem plus a real vulnerability scanner keep the agency's own stack watched.

VisualA "data boundary" wall the payroll data never crosses CTA"What would YOU need before trusting AI with this?"

Ep27

The Structure · An Agency as a System

Eight units, one nervous system

How a real government org chart becomes a system that thinks about itself.

The agency's real structure — Executive Leadership, Core Technology, Digital Infrastructure, Training, Partnerships, Media, Monitoring, Admin & Finance — is modelled in full.
The MD/CEO reports to the Governor; the CTO runs the platform; every unit has an agent that knows its work.
Because the org is data, not a slide, the Brain can answer "who owns this?" and "what's blocked in that unit?" in real time.
The org chart stops being a poster on a wall and becomes the thing the system runs on.

VisualOrg tree morphing into the live graph CTA"Could your org answer 'who owns this' instantly?"

Ep28

Roadmap · The MDA Site Generator

"Build a website" — over Telegram, in minutes

Coming next: describe an MDA over chat, get a live, reviewable website back.

Message NORA "build a site for the Ministry of Health" → a short Q&A → a complete, responsive, accessible site generated and deployed to a demo URL.
The stated colour/brand drives the whole palette — the agency's own hero videos and imagery drop straight in.
Every build parks at the gate and notifies the CTO and requester — nothing goes "official" without review.
Iterate in chat: "change the hero to X" → regenerate, redeploy the same link.

VisualA Telegram Q&A → a finished site sliding in CTA"A whole website from a chat — believe it?"

Ep29

Roadmap · Ask the Vault

Ask the agency anything, get a sourced answer

On the roadmap: a search that reads the agency's whole knowledge base for you.

Retrieval over the generated vault — every unit, policy and note the agency has written — behind a single /api/ask.
Ask a plain-language question; get an answer grounded in the agency's own documents, not the open internet.
Plus Cal.com scheduling and a stronger model tier for the heaviest work — the CTO's Hermes, Scribe's memos.
The direction: from an agency you operate to one that answers, schedules and drafts alongside you.

VisualA question → an answer with its sources cited CTA"What would you ask your whole org at once?"

Ep30

Roadmap · The Whole State

From one agency to every state employee

The big horizon: opening NORA beyond one agency — carefully, boundary intact.

A future phase: per-user identity so each staffer acts as themselves, and SSO for every state employee.
The payroll servers join only then — and only with the data boundary designed in from day one.
A unified Office surface: one inbox merging NORA, Brain messages and email; approvals as the home screen.
The ambition, stated honestly: a model other agencies — and other states — could adopt.

VisualOne agency node → a map of many lighting up CTA"Should every government run on something like this?"

Ep31

Series Finale · The Thesis

Why NORA matters

Close the loop: this was never about a chatbot. It's about how a state governs itself.

Recap the arc: a Brain that maps the agency, agents that draft, a gate that keeps a human accountable, an Office in every pocket.
The quiet radical idea: AI that amplifies public servants instead of replacing them — and never sends without a human's yes.
Built lean, built safe, built in-house — proof that world-class government tech doesn't need a world-class budget.
Credit the team and the units; invite MDAs and other agencies to reach out.

VisualThe full graph, fully lit, the tagline resolving CTA"Everything Technology, Technology Everywhere."

The Playbook

How to run the series

Cadence

One episode a day

31 posts = ~6 working weeks. Season 1 (Ep 1–19, the tour) builds the audience; Season 2 (Ep 20–31, the machine room) rewards the ones who stayed. Post in order — each references the last.

Formats

Reel · Carousel · Thread

Concept and finale episodes → short reels with the graph animating. Agent deep-dives → carousels, one card per agent. Machine-room episodes → threads or explainer reels with a diagram.

Voice

Confident, plain, human

Government-grade, no jargon dumps, no hype. Every episode ends on a human stake or a question, not a spec. Show the work; let the restraint impress.

Assets

Real screens beat mockups

Screen-record the actual force-graph, NORA voice replies, the Office push, an n8n canvas firing. The system is real — proof of that is the strongest content.

Brand

Navy, gold, teal — Georgia display

Hold the agency palette across every card so 31 posts read as one series. The tagline — "Everything Technology, Technology Everywhere" — closes recurring beats.

Extend it

Bonus & behind-the-scenes

Spin-offs when you need them: a "how we built it in-house" BTS, a myth-buster ("no, it can't send on its own"), staff spotlights pairing each person to their agent.