I built a financial analyst that lives in my Telegram and I'm slightly scared of how much I use it now

Okay so context: I'm a second-year CS student, and a few weeks ago at a hackathon I got annoyed that every "AI stock bot" I tried was either a glorified ChatGPT wrapper that hallucinated prices, or a dashboard nobody actually opens. So I built Finley instead — no dashboard, no commands, you just... talk to it in Telegram like you'd talk to an analyst friend who never sleeps. Send it a ticker, a voice note, a PDF of an earnings report, a screenshot of a chart — it pulls live data from Finnhub/yfinance/SEC EDGAR, remembers what you've asked before (actual vector memory, not just chat history), and can proactively DM you a morning briefing or a price alert without you asking. The part I'm genuinely proud of: it runs on 100% free tiers. Gemini with multi-key rotation across accounts (auto-detects rate limits, rotates keys, retries — never just dies), MongoDB + Qdrant free clusters, zero paid APIs. I wanted to prove you could build something that doesn't feel like a toy without spending a dollar. I'm posting this half-nervous, honestly — I know finance-bot posts get torn apart here (rightfully, most of them are trash), and I fully expect someone to poke a hole in the alert latency or ask why I didn't just use LangGraph. Go for it, that's kind of why I'm here. Would rather find out now than after more people are relying on it. It's open source, MIT licensed. Link's in the comments so this doesn't get auto-filtered. What would you actually want out of something like this before you'd trust it with a real watchlist?