Five real documents — technical notes, mission ToRs, budget analyses — taken from 11 hours of manual work to 1h 20min using Claude.
Cohorte Mai 2026
De l’assistant au système.
From assistant to system.
Three professionals.
Three reclaimed weeks.
Documented during the cohort on real work participants brought into the room.
Analysis, data reconstruction, budget control, and business plan work — a week of senior-leader output compressed to roughly three hours.
Three production-grade Skills replacing manual work: WhatsApp order confirmations, customer-service replies, and supplier negotiation emails.
See every documented task
| Task | Before | After Claude | Gain |
|---|---|---|---|
| Technical note — chick acquisition | 1h 30 | 15 min | 94% |
| Mission ToR — poultry farmer identification | 2h | 10 min | 92% |
| Budget analysis — Beyla unit | 4h | 30 min | 88% |
| Professional email to WFP | 30 min | 5 min | 83% |
| Steering committee minutes (COPIL) | 3h | 20 min | 89% |
| Task | Before | After Claude | Gain |
|---|---|---|---|
| Data analysis | 3 to 4 hours | 15 min | ~92% |
| Data reconstruction | 1 day | 15 min | ~97% |
| Budget controls | 2 days | 30 min | ~97% |
| Business plan | Multi-day | 1 hour | Major |
| Analysis review | 3 days | 30 min | ~97% |
| Task | Before | After Claude | Gain |
|---|---|---|---|
| WhatsApp order confirmation | Manual typing per order | Skill-driven message | Hours/day |
| Customer-service reply (SAV / complaint) | Free-form, inconsistent | Three-block structured response | Quality + speed |
| Supplier negotiation email | Ad-hoc drafting | Skill with object + body + action | Consistency |
Three modules. One arc.
Built on three frameworks established in Session 1, applied to production-grade workflows in Session 2.
The 5C Frame
Every prompt = 5 layers: Casquette · Contexte · Commande · Contraintes · Calibre.
3-Layer Memory
Personal Claude config: Profile (who I am) · Project (what I work on) · Style (how I write).
The BUILD Method
Besoin · Uploads/Instructions · Liaison/Déploiement. The assistant applied to a real task.
Skills
Your prompts become buttons.
Each prompt is rebuilt into a Skill: a rôle ancré, an explicit method, input validation, a strict output format, reference examples, explicit interdits, and a quality criterion. Seven layers that turn fragile prompts into reproducible production tools.
Connectors
Claude wired into Gmail, Drive, and Calendar.
Reading inboxes, writing into Google Docs, scheduling work — Claude operating directly on the tools each professional already uses, with the right permissions and posture.
Claude Code
Industrial automation, without writing code.
From a one-shot assistant to orchestrated, file-aware workflows: financial models, business plans, multi-document analyses — all run through Claude Code as the execution layer.
In the room.
Two Saturdays at CNPG, Conakry — May 9 and May 16, 2026.
Cohorte de juin 2026
Un métier. Un cas concret. Un assistant.
Your work. A real task. An assistant.
In person in Conakry · Online track on June 7 & 14
Finance, mining, data, reporting, commerce, administration and health: the June cohort adapts the fundamentals to different professional settings. The second session draws on participants’ practice feedback.
Structured prompts and the BUILD method
Seven professional assistant templates
Meeting minutes, quotations and presentations
Excel analysis and data reconciliation
Applied to everyday work.
Between sessions, the programme sets five uses over seven days, with a log of time spent and difficulties. Exercises include emails, meeting minutes, quotations, spreadsheets and communications.
Try it in your own work, record difficulties and improve your assistant using practical feedback.
In the room.
June cohort · Conakry, Guinea
Cohorte de juillet 2026
Mieux guider l’IA. Aller plus loin.
Guide AI better. Take it further.
Two in-person sessions · CNPG, Conakry
The third edition develops request-writing skills and makes more room for guided practice. Exercises connect Claude’s tools with office tasks and working files.
9C: clarify audience, steps, examples and checks
Memory, Projects and BUILD assistants
Skills and connectors
Demonstrations with Claude Code and Cowork
Applied to everyday work.
Demonstrations cover spreadsheet anomalies and data reconciliation. A proposed 30-day follow-up challenge includes three Skills, one connector, one Code or Cowork project and a practice log.
A practice plan to consolidate learning after training. The 30-day challenge is a proposed goal, not a measured result.
In the room.
July cohort · Conakry, Guinea
Formation CNPG 2026
L’IA au service du travail institutionnel.
AI for institutional work.
Institutional training · Conakry
Separate from the public cohorts hosted at CNPG, this programme focuses on the work of its teams: documents, finance, communications, administration and synthesis. Exercises illustrate uses adapted to institutional roles.
Financial dashboards and billing follow-up
Tender responses, procurement and contracts
Communications, archiving and synthesis
Human resources and administrative follow-up
Applied to everyday work.
The manual organises exercises around eight workplace Skills, using Word documents, Excel spreadsheets, correspondence and meeting minutes. Demonstration data is fictional or anonymised, with a human review step.
A shared approach to exploring AI across institutional roles, to be adapted and validated before operational use.
In the room.
CNPG team training · Conakry, Guinea
Bring a Rabbitron cohort
to your team.
We design cohorts for ministries, NGOs, and companies across Francophone Africa and beyond. Each cohort is built around your team’s actual work — and ships measurable outcomes.
Talk to us