Finance Data Analyst who lives in Claude Code

What I do I use Claude Code all day every day to support a finance ops team. The work involves 20,000-row sheets, Databricks SQL notebooks, Google Sheets inputs from multiple teams, covenant checks, Slack comms. Every Thursday the team produces a report that has to balance to the penny for investors. My job is to make that process smooth: I've built a Wednesday pre-check that catches problems before Thursday, reconciliation scripts that flag what's changed week-over-week, and validation that blocks the report from shipping if something's off. The setup is ~5,600 lines of Python running through Claude Code custom skills, with an Obsidian knowledge base for domain context and an LLM layer that reasons about anomalies. My background is data analysis, not software engineering. I can read SQL and Python fine, but I'm not writing it from scratch without AI. Claude Code is how I've built all of this, and it's made me significantly more technical than I was. But I know there's another level, and I want someone who's already there to help me get to it. What I need help with I want someone who uses Claude Code properly in their own work to look at my setup and help me level it up: Knowledge base structure: I have an Obsidian vault with 120+ notes, 9,000+ lines of transcripts, and domain context. But Claude still hallucinates when it doesn't have enough loaded. I need advice on the best way to structure and surface this so the model always has what it needs without me manually feeding it context. Databricks schema as persistent context: People come to me asking "why isn't this query returning X" and right now I have to copy-paste the relevant SQL/schema into the conversation every time. I want that loaded permanently. Excel input/output documentation: I work with 7-8 different spreadsheet formats weekly. My scripts handle them, but there's no clear spec of what each file is, what columns matter, how they connect. I want that written so Claude (and I) can reference it. General workflow improvement: There's a lot of copy-pasting between Slack, Google Sheets, Databricks, and local files. MCP integrations are coming soon which should help, but I want someone who can help me be ready for that and optimise what I can now. Constraints You can't access my machine or real data. Work has to be on test data or documentation. Everything outputs as Excel right now (openpyxl). Getting that right is priority before making it fancier. You You're a data analyst who uses Claude Code as your primary tool. Not someone who's "tried it out" or "familiar with AI tools." Someone who runs their actual job through it: queries, scripts, knowledge bases, the lot. In your application, show me examples of how you use Claude Code to supercharge your data analysis. What have you built? How do you structure context so the model doesn't lose the plot? What's your setup look like? Python, pandas, SQL comfortable Can look at someone else's setup and tell them what they're doing wrong Understand knowledge base design (how to stop LLMs hallucinating with the right context structure) UK timezone or close Short engagement to start, could extend.