MaskMind: Fine-Tuned Financial Mask-Filler Model

• GIF

  Quick demo of the project running

Inspiration

Financial reports are dense, repetitive, and hard to digest — especially SEC 10-K filings. I wanted to make reading financial data faster, smarter, and more interactive by fine-tuning a language model specifically on real financial documents.

What It Does

MaskMind: Financial Mask-Filler MLM allows users to enter partial financial sentences with [MASK] tokens, and the model intelligently predicts the missing information based on Apple's SEC 10-K filings.

It’s like "autocomplete" — but for financial experts.

How I Built It

• Data Collection: Downloaded Apple's SEC 10-K annual reports
• Preprocessing: Cleaned, tokenized, and chunked the raw text
• Model Fine-Tuning: Started with BERT (fine-tuned on SQuAD), then custom fine-tuned further using masked language modeling (MLM) on Apple's financials
• Gradio App: Built a secure, private app that users can interact with directly without needing an internet connection

Challenges I Faced

• Dealing with the size and complexity of 10-K filings
• Avoiding generic predictions by re-training on specialized domain-specific financial language
• Hosting large model checkpoints outside GitHub without losing accessibility

Accomplishments I'm Proud Of

• Built a fully private, secure financial domain MLM
• Trained a fine-tuned masked language model from scratch
• Created a professional, clean Gradio web interface ready for real-world use

What's Next

• Expand to other companies beyond Apple
• Add multi-company or industry-specific models
• Build lightweight financial Q&A systems alongside the mask-filler

Built With

• 10-k
• 3.12
• apple
• custom
• docker-ready
• filings
• fine-tuning
• gradio
• huggingface
• python
• scripts
• sec
• transformers