The Challenge

When we first met the team at Fyler.us, they were struggling with the manual effort required for 8-K filings. These critical reports could take up to 12 hours to prepare—especially under tight deadlines for material event disclosures. The process was not only time-consuming but also prone to formatting errors that could cause delays.

Recognizing the potential of AI in regulatory compliance, CEO Cory Skerl set out to modernize the filing process and eliminate the risk of last-minute scrambles. Enter the M3S team, whose expertise with large language models (LLMs) and retrieval-augmented generation (RAG) fundamentally changed Fyler’s game plan.

The Catalyst

To see which model—or combination of techniques—could best manage real-time news updates, legal references, and the nuanced language of compliance, the M3S team orchestrated a series of rigorous, double-blind trials. They tested multiple LLMs ranging from OpenAI o1 to Llama-3.1 variants, comparing accuracy, style, and cost.

Our Approach

• Double-Blind Trials Across LLMs: M3S compared OpenAI o1, Llama-3.1 (405b), and Llama-3.1 (70b) to see which handled 8-K disclosures most effectively.
• RAG Integration: By pairing language models with retrieval-augmented generation, their system pulled precise language from prior filings to reduce errors and missed disclosures.
• Splitting Item Identification from Full-Text Generation: This two-part process boosted both accuracy and consistency, ensuring each disclosure item was accurately flagged before final text was generated.
• Flexible, Modular Pipeline: Automatically aggregated data from PDFs, existing 8-Ks, and real-time news feeds, then funneled it into the LLM. This design allowed quick adjustments as new regulatory demands or model updates arose.

Additional Solution Highlights

Beyond experimenting with multiple LLMs, the M3S team also:

• Built and trained a custom AI pipeline using thousands of existing filings
• Implemented automated EDGAR compliance checks
• Developed a system to auto-generate draft disclosures
• Created an intuitive interface for quick reviews and edits

The Results

With OpenAI o1 emerging as a standout performer for correctness and style, and RAG integrations fueling precise language pulls from real filings, the approach slashed Fyler.us’s 8-K prep times and minimized errors. The integrated EDGAR checks also maintained perfect compliance on first submission. Meanwhile, the M3S team’s vision for a specialized FylerLLM offered a tantalizing glimpse into fine-tuned models for 10-K and S-1, although that initiative was paused due to compute constraints.

Looking Ahead

By merging AI innovation with a data-centric approach, Fyler.us has emerged as a frontrunner in regulatory compliance automation. The M3S team tackled the immediate pain points of 8-K filings while laying out a strategic path for large-scale AI adoption.

This transformation—from 12-hour all-nighters to sub-hour efficiency—illustrates the power of modern AI-driven document generation. For organizations grappling with high-stakes reporting and evolving regulations, Fyler’s success story stands as proof that the right blend of innovation and strategic foresight can revolutionize compliance.