Powered by Qwen2-0.5B · 100% Local Inference

Decode SEC Filings
with Local AI

Enterprise-grade financial intelligence. Upload 10-K, 10-Q, and 8-K filings or pull directly from EDGAR. Every byte stays on your machine.

10-K Annual Reports
10-Q Quarterly Filings
8-K Current Reports
0 API Calls to OpenAI
filinglens.local
$ filinglens analyze --ticker AAPL --type 10-K
Fetching latest 10-K from EDGAR…
Extracted 94,217 words
Splitting into 28 chunks…
Running Qwen2-0.5B locally…
Sentiment: Positive (82% confidence)
Analysis complete in 48.3s
CAPABILITIES

Built for Financial Analysts

No cloud. No subscriptions. No data leaks. Just raw financial intelligence running on your hardware.

Local LLM Inference

Qwen2-0.5B-Instruct runs entirely on your machine via HuggingFace Transformers. Zero data leaves your network.

On-device AI

EDGAR Integration

Fetch 10-K, 10-Q, and 8-K filings directly from SEC EDGAR using any ticker symbol. No manual downloads needed.

Auto-fetch

Smart Chunking Pipeline

Proprietary chunking engine splits 100K+ word filings into overlapping segments, processes each, then aggregates results.

Long-doc support

Sentiment Distribution

Visual breakdown of Positive/Neutral/Negative sentiment across all analyzed chunks with majority-vote aggregation.

Visual analytics

Section Targeting

Focus analysis on specific filing sections — Risk Factors, MD&A, Liquidity — for surgical financial review.

Precision analysis

Drag & Drop Upload

Upload .txt, .html, or .htm SEC filings directly. HTML is automatically cleaned and normalized before processing.

Multi-format
ANALYSIS ENGINE

Run Your Analysis

Upload a filing or pull one from EDGAR. The local AI will handle the rest.

Drop your SEC filing here

Supports .txt · .html · .htm · Max 50 MB

Leave blank to analyze the entire filing

01 Extract & Clean Text
02 Chunk Document
03 Local LLM Inference
04 Aggregate Results

Ready to Analyze

Submit a filing to see sentiment analysis, chunk breakdown, and financial insights appear here.

ARCHITECTURE

How It Works

FilingLens AI uses a multi-stage pipeline to handle SEC filings that often exceed 200,000 words — far beyond any single LLM context window.

1

Ingest

Files are uploaded or fetched from EDGAR. HTML is parsed and stripped to clean text using BeautifulSoup.

2

Chunk

Documents are split into 800-word overlapping segments. The first and last halves are prioritized for very long filings.

3

Infer

Qwen2-0.5B-Instruct analyzes each chunk independently with a financially-tuned prompt. Runs 100% locally via PyTorch.

4

Aggregate

Majority voting across chunks produces a filing-level verdict. Sentiment distribution is computed and visualized.

System Specs

Model Qwen/Qwen2-0.5B-Instruct
Parameters 500 M
Chunk Size 800 words
Max Chunks 40 per filing
Backend Flask + Python 3.10+
Inference PyTorch · device_map=auto
Data Policy 100% local · no cloud