chinmaymk

I recently tried to use GPT Astra and Anthropic models to see if they decipher the indus valley script, long story short - they didn't. I burned ~891M tokens on this across 3 weeks and several resets. The entire exploration is captured in a github repository with research log wrt directions and experiments conducted.

Some background on Indus valley script: The Indus script is the undeciphered system of signs left by the Indus Valley, or Harappan, civilization, which flourished in ancient Indian sub continent. Its signs were widely used around 2600–1900 BCE, including at Harappa and Mohenjo-daro.

There's a $1M award for decipherment. https://www.cnn.com/2025/02/28/asia/indus-valley-script-indi...

I built ra to understand how tools like Claude Code and Codex actually work under the hood. It's a single CLI binary where a config file is the agent — change the config, change what the agent does. Same binary runs as a REPL, HTTP server, cron job, or MCP server. Drop a config in your repo and your whole team gets the same agent.

Every line of code, tests, docs was written by AI using Claude Code. The blog post covers the full workflow — what worked, what didn't. If you need a simple, cross-provider agent harness, give it a shot!

https://github.com/chinmaymk/ra

Oh, didn't know this earlier, thanks for the info!