Temperature and rainfall visualizations for Bangalore (Bengaluru), using hourly weather data from 1981 to present via the Oikolab API.

bangalore_weather_update.R -Downloads the latest hourly temperature and precipitation data from Oikolab, appends it to the local .RData files, and produces a chart for the current year. Includes a 3-bullet AI commentary summarising the last 2 weeks of weather compared to historical norms. Skips the API fetch gracefully if data is already up to date.

bangalore_weather_historical.R -Generates year-by-year weather charts for any past year. Each chart includes:

• Temperature panel: Daily high/low range for the target year (brown), overlaid on the normal range (grey) and all-time record range (beige). Days that broke historical records are labelled.
• Rainfall panel: Cumulative monthly precipitation (brown bars) compared to the historical monthly average (green step line). Record rainfall days and wettest months are labelled.
• AI commentary: 4 bullet points highlighting the most notable weather patterns for that year, generated by Claude (Haiku) from pre-computed statistics.

The commentary works by having R compute ~30 weather statistics (hot/cold day counts, streak lengths, rainfall totals, etc.), rank them by how far they deviate from historical averages, and send the top 5 to Claude Haiku for concise rephrasing. The LLM's role is minimal - it synthesises pre-selected facts into readable bullets rather than doing any analysis.

Temperature thresholds for "hot days", "cold days", etc. are data-driven (historical percentiles), not hardcoded.

Both scripts resolve their own directory at runtime (sys.frame(1)$ofile when source()'d, --file= when run via Rscript, rstudioapi in RStudio), so they work regardless of working directory.

Static HTML artifacts intended for browser viewing can be published from the docs/ folder with GitHub Pages.

• Landing page: docs/index.html
• Few-shot review cards: docs/fewshot_annotations/review_cards.html

Use analysis_questions/question_template.Rmd when a new weather question needs a notebook-backed answer. If the result is genuinely worth keeping, write a blog post in docs/blog/, add it to the top of docs/blog/index.html and docs/blog/feed.xml, add it to the top of analysis_questions/analyses.yml with insightful: true, publish: true, and a blog_url, then run:

Rscript analysis_questions/update_site_analyses.R

That regenerates the latest three homepage blog cards in docs/index.html using the same article format as the existing site. Published, insightful entries must point to a blog post.

Blog posts live in docs/blog/. Each post must include the visible disclosure This post is AI-written. at the top of the article body. The updater fails if a published analysis is missing its post, disclosure, or RSS entry.

• R with packages: tidytable, tidyverse, patchwork, lubridate, ggthemes, ggrepel, httr2, jsonlite, yaml, rmarkdown
• An Anthropic API key in ~/.Renviron as ANTHROPIC_API_KEY (for AI commentary; charts render fine without it)
• For bangalore_weather_update.R: an Oikolab API key in ~/.Renviron

# Generate a single year's chart
source("bangalore_weather_historical.R")
generate_weather_chart(2024, save_path = "bangalore_weather_2024.png")

# Generate multiple years from the command line
Rscript bangalore_weather_historical.R 2021 2022 2023 2024 2025

The underlying hourly data (bangaloreTemperature.RData, bangaloreRainfall.RData) is not included in this repo due to size. Run bangalore_weather_update.R to fetch and build the data files.

Chart design inspired by the NYT/Tufte style -minimal gridlines, muted earth-tone palette, data-ink focused. Uses ggthemes::theme_tufte() as a base.