bongo.ai hex stickerbongo.aiHack Your Data

Before the week

Install the toolkit

Before Monday

Install the toolkit once, on your own machine or in the cloud. Monday morning is for thinking, not drivers.

You will leave able to

  • Run R and Posit (RStudio) Desktop, or work entirely in Posit Cloud if IT blocks installs.
  • Install the tidyverse, tidytext and Quarto.
  • Set up an AI copilot (bongo.ai in this course; ChatGPT / Claude / Copilot at work) without pasting confidential data into it.

On the desk

Install R

R is the language. Posit Desktop (still widely called RStudio) is the workbench. Install R first, then the workbench. Download the current release for your operating system from CRAN.

  • Windows. Choose *base*, then install RTools so packages that need a compiler can build.
  • macOS. Install the latest .pkg. On Apple silicon, prefer the arm64 build. XQuartz is only required for a handful of older graphics packages.
  • Linux. Use your distribution’s CRAN binary instructions rather than compiling from source.

Install Posit Desktop

From posit.co/download take the open-source Desktop installer for your platform. After first launch: *File → New Project → New Directory → New Project*. Name it hack-your-data. All course files live inside that project so paths do not break when you move laptops.

Posit Cloud if IT says no

Many banks and FMCG firms lock down local installs. Posit Cloud runs the same IDE in the browser. Create a free account, start a New Project, and work from there. Do not upload payroll, customer-level KYC, or unpublished results. Use the anonymised course extracts, or a synthetic twin of your own data.

Packages you will actually use

R
install.packages(c(
  "tidyverse",  # readr, dplyr, tidyr, ggplot2, purrr
  "tidytext",   # one-token-per-row text
  "gt",         # publication tables
  "scales",     # NGN, %, dates on axes
  "janitor"     # clean names
))
 
# Quarto is a separate install: https://quarto.org
# In the terminal:
# quarto --version
Run once per machine. Then never install inside a report.

AI tools — with a data-handling protocol

This programme treats large language models as copilots, not oracles. You will use the in-site bongo.ai panel on every session page. At work you may also have Microsoft Copilot, ChatGPT Enterprise, or a private Azure OpenAI endpoint.

  1. Never paste a customer name, BVN, account number, or unpublished earnings figure into a consumer model.
  2. Strip columns to the minimum the prompt needs. Prefer *schema + three fake rows* over the real extract.
  3. Ask the model for code, not for the answer. Then run the code on your machine.
  4. Read every generated line. If you cannot explain it to a colleague, you cannot put it in a pack.

A 60-second smoke test

R
library(tidyverse)
 
tibble(
  channel = c("mobile", "branch", "agent"),
  nps = c(58, 42, 47)
) %>%
  ggplot(aes(channel, nps)) +
  geom_col() +
  labs(title = "If you can see this, you are ready for Monday")

Next: Computational thinking