I just spent 3 weeks building this. Our new multi-agent architecture is live. We now support: • 10 LLMs • 15 agents • 20,000+ stocks LangChain orchestrates our agent graph. All code is open source.

I was not expecting this. We had 5 LLMs analyze stocks for red flags. Good: LLMs caught 100% of risky stocks ✅ Bad: 25% of safe stocks were flagged ❌ Best performers: • opus • o3 • gemini 2.5 pro My takeaway: LLMs excel at spotting risk in stocks, but may be too cautious

I’ve been putting this off for weeks. Extracting financials from SEC filings is hard. The hardest part is ensuring extracted data is clean. Doing that in real-time is even harder. But LLMs need fresh data, not yesterday’s numbers. So today, I’m finally tackling it. If this

Extracting financials from SEC filings is hard. Every company reports things differently. The real challenge is standardizing it all. If you’re lucky, values like cost of revenue are reported directly. Usually, you have to calculate them yourself. And in the worst case,

I’m starting a new project this week. An open source finance web crawler. We’ll scrape: • earnings transcripts • market news • global stock data The idea is to scrape hard-to-find data that doesn’t live in clean APIs. I’ll then layer a simple API in Financial Datasets Time to

Extracting financials from SEC filings is hard. Every company reports things differently. The hardest part is standardizing the data. Line items like “cost of revenue” often aren’t reported directly. So you have to figure it out yourself across 17,000+ possible tags. For

This one’s been on my list forever. My segmented revenue API will soon span 20,000 tickers and 15+ years. It lets your LLM see how a company makes money. In real-time, when earnings drop. Revenue breakdowns by: • geography (US vs. China) • product line (iPhone vs. iCloud)

GPT-5 is live in our AI hedge fund! I added all 3 models: • gpt-5 • gpt-5 mini • gpt-5 nano Can't wait to test them out.

No one’s pulled this off yet. A stock market API that updates the moment earnings drop. I’ve built this for months. It’s almost ready. Yesterday I tested it on $ABNB. The Q2 2025 income statement hit our API 3 seconds after release. Data was clean. Structured. Accurate.

I’ve been testing open source LLMs lately. Just added gpt-oss to our AI hedge fund. Both sizes: • gpt-oss-20b • gpt-oss-120b Pulled and running locally with ollama in minutes.

I ran some numbers this morning. Yesterday financial datasets served: • 1.21M requests • 183ms p95 latency • 0 total errors This is the kind of stability I want every customer to feel. I’ll be writing long-form posts soon on the performance tweaks behind the API.

SEC filings now hit Financial Datasets ~15s faster. We used to poll the SEC every 30s, but the bottleneck was wasted time in each cycle. Fixed it by: • Reusing HTTP connections • Parallelizing network I/O to EDGAR • Eliminating redundant DB writes in hot path Result: Same

It’s time to evolve our AI hedge fund. Instead of 1 team of agents, we'll have multiple “pods” - each with analysts and PM. The pods report to a Chief Investment Officer agent, like a real hedge fund. We’ll use LangChain to run the multi-agent pods in the background. The

I promised I’d share this. How I wired OAuth into my MCP server. There are 3 players: • MCP client (Claude, etc.) • MCP server (@cloudflare) • Backend (Financial Datasets) The tricky part: each party needs to trust the other without sharing ANY secrets. I used OAuth 2.1 and a

This is a wonderful feeling. I’ve been heads down on performance work for Financial Datasets all week. Last 24 hours: • 1.34M requests served • 95% finished in ~200ms • 0 new issues Why I care: LLM-powered financial agents already have compute overhead. Real-time stock market

Big week for our AI hedge fund. We added new investor agents and LLMs. Current state: • 15 research agents • 10 local LLMs via ollama • 20 hosted LLMs Next up: multi-agent analyst pods with LangChain + LangGraph. Everything is open source. No coding experience

It’s finally live. SEC filings on Financial Datasets MCP server. Adds to existing tools: • get news • get stock prices • get financials, etc. Our filings are real-time, as they hit the SEC.

I upgraded our AI hedge fund’s risk manager. We now check correlation between every stock we’re considering. Why? So we don’t load up on companies that move together. New logic: • High correlation → smaller position • Low correlation → slightly bigger position Combined

Another record day. Financial Datasets served 1.3M requests in 24h. My goal: disrupt legacy finance and make high quality stock market data accessible. Still early. The journey continues. Grateful for the support ❤️

I just upgraded our Valuation agent. The DCF model was too basic, until now. Here’s what’s new: • smarter WACC calculation • multi-stage cash flow models • market regime analysis (bear / bull / neutral) Feels good to see this agent evolve. It’s one of the most popular ones