AI-Driven Investing Research Platform
The Invest Collective
The Challenge
The Invest Collective is a private investing group whose members generate far more rigorous research than the tools they used to coordinate it. Investment theses lived in text threads, spreadsheets, and Google Docs. Members tracked their own data points by hand, and the group's collective track record -- the reason anyone joined in the first place -- was effectively unpublishable because no two sources agreed on what a thesis had actually claimed, when, or how it had resolved.
The natural fix -- a real-time research platform with market data updates and a published track record -- collided with a hard budget reality. A conventional build would layer a Bloomberg or equivalent data feed, an Airflow ETL tier, and a cloud data warehouse underneath a web application. That stack runs $500–$5,000 per month in recurring infrastructure before a single feature ships. For a private group operating at a personal scale, that cost structure was a non-starter.
Our Approach
We designed a members platform that treats the group's discipline -- testable predictions, bounded timeframes, measurable data points, and a public outcome score -- as the product itself. The Thesis Tracker became the centerpiece. Every thesis ships with 2–8 data points, a target value and direction for each, and a confidence score that recalculates on every update. When the thesis closes, a permanent, stricter outcome score replaces it. Wins and losses are both visible, because a track record is only credible if the misses are on the same page as the hits.
Underneath that product, we replaced the classical market-data stack with an AI workflow. A small stateless MCP server -- roughly 600 lines of TypeScript -- gives Claude a Zod-validated set of tools: search recent financial news, draft a thesis, update a data point, close a thesis. Every tool call writes through the same production API that humans use. The API owns truth and validation; the AI is a draftsman, not an oracle. A nightly scheduled task, running on Anthropic's cloud rather than anyone's laptop, walks every active thesis after market close, pulls current values from Yahoo Finance, FRED, and MarketWatch, updates the database, recomputes scores, and closes any thesis whose end date has passed. Humans review every AI-drafted thesis before it goes public.
The rest of the platform was built around the same principle of radical simplification. The research library lets admins upload PDFs to object storage and optionally auto-summarize and auto-tag with edge-hosted inference. A cookie-free analytics layer writes one row per pageview to a SQLite table at the edge and renders six dashboard views from fourteen aggregated queries batched into a single round-trip. The admin panel sits behind Cloudflare Access, with Google, GitHub, or OTP identity and no custom login code, password reset flow, or session store to maintain. API endpoints use a separate Bearer-token scheme so external systems -- including the MCP server running outside the perimeter -- can write to the platform without touching browser auth.
The entire platform -- public site, admin panel, API routes, analytics ingestion, and AI calls -- deploys as a single Cloudflare Worker. Page rendering, the SQLite database, object storage, and model inference all live at the edge, under one deploy target with one monthly bill.
The Results
The MVP shipped with four working systems: a members homepage and application flow, an AI-assisted research library with PDF support, the public Thesis Tracker with live scoring, and a privacy-clean analytics dashboard. Infrastructure cost landed at effectively zero -- the platform runs inside an existing Cloudflare Workers plan shared across multiple sites -- against a classical equivalent that would have required roughly fifteen discrete services and $500–$5,000 per month in recurring spend.
Beyond the headline numbers, the engagement proved a thesis of its own: a small team, paired with a disciplined AI workflow, can deliver a product that a few years ago would have required a data engineering org. The Thesis Tracker turned an informal group of serious investors into something publishable. The nightly scheduled job removed the last human bottleneck from the data pipeline. And the Cloudflare-native architecture left the client with a stack they can extend without hiring around it.
This outcome was made possible by the combination of Cloudflare's edge computing capabilities, the power of AI-driven workflows, and a deliberate focus on simplification, resulting in a platform that is not only faster and more cost-effective but also more secure and easier to maintain.