hi, i'm ani.
previously worked on real-time agent i/o at structured ai (F25 ) and our bad habit, an atlantic records venture. every now and then i post about what i'm doing with claude code and codex, as featured on business insider.
structured ai drawing chat with page-level citations architectural PDFs in, cited answers out. streamed claude/gemini and kept redis replay for live spectating and later debugging. quantercise quant prep with real grading next.js, typescript, postgres, drizzle, stripe, sandboxed python, and math-heavy grading paths. paragon global investments research portal for a quant fund next.js and typescript on supabase. made fund research searchable and usable from mobile instead of buried in scattered docs. public tooling tools i actually use claude-code-tips and imessage mcp are small public receipts from the same local-first workflow i run every day.
making
view allwriting
view all saturdays are for claude code business insider interviewed me about ai usage limits. the useful part was less the quote and more the workflow it forced. i built a monitor for my claude code sessions claude code has no dashboard for parallel sessions, so i'm building claudemon. stop ending your day with "fix the bug" vague todos waste context. specific prompts let claude code start from the right file.
newsletter
subscribe notes on agent workflows and product builds, including the parts that broke while shipping.
