1. One-Click End-to-End Workflow Pipeline & Technical Architecture
5-Stage Pipeline
Enter any target company (e.g. ASML, TSMC, NVDA, NXP, VSH, AMAT) or a full CompaniesMarketCap URL, choose the number of years (3 to 10), and click "Run End-to-End Workflow". The system executes a fully automated 5-stage pipeline:
1. Crawl & Cache
2. PDF to MD Tables
3. Dual-Track Extraction
4. OpEx KPI Engine
5. Visual & LLM Synthesis
1
Stage 1: Target Resolution & Sub-Second Cached Crawler
10-K / 20-F / 10-Q
Local Cache
Automatically detects issuer domicile (US Domestic Form 10-K/10-Q vs. Foreign Private Issuer Form 20-F/6-K) and target reporting horizon. Downloads official SEC PDFs directly into data/downloads/. Incorporates high-efficiency local caching so previously downloaded filings load in 0.1s without re-requesting the web.
2
Stage 2: Structural PDF-to-Markdown Lossless Table Parser
Table Grid Scanner
Item 8 / Item 18
Solves the fundamental problem where traditional PDF converters scramble multi-column table layouts. Uses coordinate grid table extraction to pinpoint audited statements (Income Statements, Segment Revenue, and Headcount sections), outputting clean, standardized GitHub Markdown tables (.md) in data/parsed_md/.
3
Stage 3: Dual-Track Financial Extraction & 10-Q Linear Interpolation
Track A + Track B
Quarterly Interpolation
Track A (Benchmark Library): Instant zero-delay loading for audited benchmarks (ASML, TSMC, NVDA, NXP, VSH, AMD, AMAT, GOOGL, etc.).
Track B (LLM Semantic Deduction): For new companies, derives key line items (Revenue, COGS, Gross Profit, OpIncome, R&D).
10-Q Headcount Interpolation: Because SEC Form 10-Q does not mandate quarterly employee disclosures, the engine anchors to annual 10-K audit numbers and computes smooth linear quarterly headcount interpolation ($Q1 \rightarrow Q2 \rightarrow Q3 \rightarrow Q4$).
4
Stage 4: Strategic OpEx & Human Capital Productivity Engine
Rev/GP/OI per FTE
The Pivot (人力拐點)
Synthesizes financial statements with organizational workforce scale. Computes the Productivity Trio (Revenue per FTE, Gross Profit per FTE, Operating Income per FTE), evaluates Operating Leverage (fixed cost absorption), calculates R&D Reinvestment Intensity %, and identifies "The Pivot" (when gross margins expand via automation even after headcount plateaus).
5
Stage 5: Interactive Dual-View Dashboard & LLM Executive Synthesis
Plotly Charts
16:9 Presentation Prompt
Renders 6 interactive Plotly visual charts with HD zoom modal and CSV exports across both Single Company Deep Dive and Multi-Company Peer Comparison modes. Enables 1-click Markdown copying to feed into Gemini / Claude / ChatGPT with project prompts (fininacial_prompt.md) to generate 16:9 C-suite presentation slides in seconds.
2. Multi-Company Peer Comparison Mode
Switch between "Single Company Deep Dive" and "Multi-Company Peer Comparison" at the top. In comparison mode, check multiple companies to analyze cross-company Gross Margin pricing power, Human Capital Productivity ROI ($/FTE), Operating Leverage, R&D Reinvestment Intensity, and Normalized % Cumulative Stock Returns side-by-side.
3. Top Switcher vs. Bottom Console (Two-Way Synchronization)
Top-Right Dropdown (Company Switcher)
Use this to instantly view and toggle between already-processed companies in 0.1 seconds without re-downloading.
Bottom Console (Pipeline Console)
Use this to download and process new companies or change the target history length (e.g. 10 years).
4. Visual Charts & Strategic OpEx Framework Guide
• Chart 1: The Pivot (Headcount vs GM %) — Identifies when workforce expansion plateaus and margin expansion shifts to automated operational excellence.
• Chart 2: Productivity Trio (Rev / GP / OpIncome per FTE) — Quantifies human capital efficiency and automation return on investment (ROI).
• Chart 3: Operating Leverage (OpIncome & Margins) — Evaluates fixed-cost absorption and profitability resilience across cyclical demand swings.
• Chart 4: R&D Commitment (Expense & % of Revenue) — Analyzes technological differentiation and moat reinvestment intensity.
• Chart 5: Growth Dynamics (YoY Triangulation) — Compares Revenue YoY, GP YoY, and OpIncome YoY against Headcount growth to ensure profitable expansion.
• Chart 6: Value vs. Volume Paradox — Disaggregates product mix between high-margin flagship lines and physical factory throughput volume.
• Chart 7 (Stock Benchmark): Multi-Company Cumulative Return (% Return) — Normalizes peer equity performance to 0.00% at the start of chosen timeframe ($R_t = (P_t / P_0 - 1) \times 100\%$), eliminating currency (USD/TWD/JPY/KRW) and nominal price gaps to track true relative alpha.
5. Gross Margin vs. Operating Margin: Core Principles & Distinctions
Gross margin measures profit after deducting only the direct production costs, while operating margin measures profit after deducting both direct costs and regular operating expenses.
| Metric |
Gross Margin |
Operating Margin |
| Formula |
((Revenue - COGS) / Revenue) × 100% |
(Operating Income (EBIT) / Revenue) × 100% |
| Costs Deducted |
Only COGS (raw materials, direct factory labor, direct manufacturing overhead) |
COGS + Operating Expenses (OPEX) (SG&A, R&D, marketing, sales commissions, depreciation) |
| Costs Excluded |
OPEX, non-operating expenses, interest, taxes |
Non-operating expenses, interest, taxes |
| What It Measures |
Production & pricing efficiency: how cheaply a company produces a good or service relative to selling price |
Overall business operational efficiency: how well management runs daily core operations |
| Typical Level |
Higher (typically top of the income statement) |
Lower (sits below gross profit on the income statement) |
• Gross Margin in practice: If a company makes $100 in revenue and spends $40 on direct manufacturing costs (COGS), its gross profit is $60, giving a 60% gross margin.
• Operating Margin in practice: If that same company also spends $30 on administrative salaries, office rent, and marketing (OPEX), its operating profit (EBIT) is $30, giving a 30% operating margin.
Neither metric includes debt interest or corporate taxes—those are deducted further down the income statement to arrive at Net Profit Margin.
5. Multi-Format Global Filing Systems & Accounting Frameworks Comparison
10-K / 10-Q / 20-F / TWSE / Yuho
Global tech enterprises file statutory disclosures under different jurisdictions, accounting standards (US GAAP vs. IFRS), and reporting frequencies:
| Filing Type |
Jurisdiction & Benchmark Issuers |
Frequency |
Audit Status |
Headcount Disclosure (員工人數) |
Key Contents & Structure (核心內容架構) |
| Form 10-K |
US Domestic (SEC)
(NVDA, AAPL, MSFT, NXP, AMAT) |
Annual
(年度) |
Audited
(查核簽證) |
Mandatory (Item 1 / 6)
Official year-end employee total |
Item 8 audited financial statements, Item 7 MD&A, segment sales, human capital |
| Form 10-Q |
US Domestic (SEC)
(NVDA, AAPL, MSFT, NXP, AMAT) |
Quarterly
(Q1, Q2, Q3) |
Unaudited
(期中未審計) |
Optional
(Linear interpolation from 10-K) |
Quarterly P&L, balance sheet, MD&A, segment sales (Q4 reported in 10-K) |
| Form 20-F |
Foreign Private Issuers (FPI)
(ASML - Netherlands, TSMC ADR) |
Annual
(年度) |
Audited
(查核簽證) |
Mandatory (Item 6.D)
Functional & regional headcount |
Item 3.A 5-year selected financial data, Item 18 IFRS/US GAAP statements |
| Form 6-K / 10-F |
Foreign Private Issuers (FPI)
(TSMC, ASML, ASE) |
Interim / Event
(季度/重大訊息) |
Varies
(依母國法規) |
Aligned with 20-F
(Quarterly anchor) |
Foreign quarterly earnings releases, material events, dividend announcements |
| TWSE Annual (台灣年報) |
Taiwan Listed Companies (TWSE)
(Foxconn 2317, TSMC 2330, ASE 3711) |
Annual
(年度) |
Audited
(會計師簽證) |
Mandatory
Parent & Global Consolidated Group FTE |
Page 5 Table 5.1 5-year financial data & profitability analysis, 4 product groups |
| Japan Yuho / Korea DART |
Japan / Korea Exchanges
(Advantest 6857, Samsung 005930) |
Annual / Qtr
(年度/季度) |
Audited
(查核簽證) |
Mandatory
Consolidated Workforce |
J-GAAP / IFRS financial statements, division revenue (Memory / Foundry / ATE) |
• Why does Form 10-Q lack Headcount? SEC Rule 13a-13 does not require quarterly headcount disclosures. Our engine automatically anchors to official 10-K/20-F annual audit numbers and performs linear interpolation ($Q1 \rightarrow Q2 \rightarrow Q3 \rightarrow Q4$) to keep productivity ($/FTE) accurate.
• Taiwan TWSE Annual & Global Group FTE: Taiwan corporate annual reports (Hon Hai / Foxconn) report parent workforce (~4,000) and consolidated global manufacturing group workforce (~650,000-850,000 FTE). The engine parses consolidated figures to maintain cross-border productivity parity.
6. Multi-Currency & Unit Scale Automatic Normalization Engine (USD $M Standardization)
USD ($M) Normalized
To enable true apple-to-apple cross-company and cross-border comparisons (e.g. TSMC vs. ASML vs. NVDA vs. Foxconn), our OpEx engine automatically standardizes all reported figures into USD Millions ($M) and applies rigorous scale sanitization:
1. Historical Benchmark Exchange Rate Matrix
FX Rates
Converts international filings using official historical annual & quarterly benchmark exchange rates:
- TWD (NT$): 29.5 (2020), 28.0 (2021), 29.8 (2022), 31.1 (2023), 32.0 (2024–25)
- EUR (€): 1.14 (2020), 1.18 (2021), 1.05 (2022), 1.08 (2023–25)
- JPY (¥): 106.8 (2020), 109.8 (2021), 131.5 (2022), 140.5 (2023), 151.0 (2024–25)
- KRW (₩): 1180 (2020), 1145 (2021), 1290 (2022), 1305 (2023), 1360 (2024–25)
2. Scale Sanitization & Financial Sanity Checks
Auto Guardrails
Automatically detects reporting units in table headers to eliminate 1,000× or 1,000,000× scale distortion:
- In Thousands (千元): Multiplies by $10^{-3}$ to normalize into $M (e.g. Palantir).
- In Millions (百萬元): Direct 1.0 multiplier (Standard SEC & Taiwan TWSE).
- In Billions / Trillions (十億/兆元): Multiplies by $10^3$ (e.g. Foxconn NT$6.86T).
- Sanity Guardrail: Verifies $GP \le Revenue$ and $GM\% \in [0\%, 100\%]$ to safeguard $/FTE metrics.
7. Why 20-F / 10-K / 10-Q Parsing Still Needs LLM Semantic Dynamic Correction
Dual-Track Engine
While our parser structurally converts 20-F, 10-K, and 10-Q tables into flawless Markdown, real-world corporate SEC filings exhibit non-standard line items and unstructured narratives that require LLM (Gemini / Claude / GPT) semantic deduction:
| Challenge Scenario |
Rule / Regex Bottleneck |
LLM Semantic Dynamic Solution |
| 1. Non-Standard Accounting Items |
Google & Meta omit the Gross Profit line, reporting only Revenues and Cost of revenues. Pure regex returns 0% GM. |
Deduces accounting relations automatically: Gross Profit = Revenues - Cost of revenues. |
| 2. Unstructured Headcount Text |
NVIDIA, AMD bury workforce totals in natural language prose across Item 1 / Item 6 without standard table grids. |
Understands sentence semantics to isolate total employees, R&D staff, and factory personnel accurately. |
| 3. Dynamic Segment Sales Breakdown |
Each company has completely unique business segments (Google: Search/Cloud; Amazon: AWS/Retail; ASML: EUV/DUV). |
Parses Segment Notes to disaggregate multi-tier product revenue and map to sales_breakdown. |
| 4. Unit Scale & Currency Traps |
Filings mix in thousands (e.g. Palantir), in millions, or foreign currencies (JPY, TWD, EUR). |
Performs financial sanity checks against table headers to prevent 1,000× scale distortion in Productivity ($/FTE). |
| 5. OpEx Strategy & Lean Maturity |
Qualitative management commentary in Item 7 (MD&A) cannot be parsed by numerical regular expressions. |
Evaluates qualitative manufacturing reforms, automation progress, and generates 5-Stage Lean Maturity ratings. |
• The Dual-Track Synergy: PDF-to-Markdown Parser handles structural table fidelity; LLM acts as the senior financial analyst brain to dynamically infer non-standard metrics, ensuring 100% universal adaptability for any global enterprise.
8. Using Parsed Markdown with LLMs (Gemini / Claude / ChatGPT)
Select any parsed .md file in the bottom browser, click "Copy Markdown", and paste it into Gemini with fininacial_prompt.md or sale_breakdown.md for instant 16:9 executive presentation decks and pitch scripts.
9. Zero-Cost Universal Access, No API Key Required & Multi-Agent Dispatch Guide
100% Free & Open
情境 A：一般使用者 / 團隊主管 / 外部訪客
100% 免 AI、免 API Key、免花費、免安裝！
• 線上直接用：點擊 GitHub Pages 網址，瀏覽器直接渲染 62 家科技巨頭財報指標、6 大圖表與 7 大賽道研究筆記。
• 完全離線單機版：將 standalone_dashboard.html 複製到任何電腦，點擊兩下即可在 Chrome/Edge 離線秒開。
情境 B：進階使用者 / 想要新增新公司財報
兩套自動化軌道自主選擇：
• 軌道 1 (免 API / 本地規則管線)：執行 python update_pipeline.py --ingest-pdf <path> --ticker <sym> --year <YYYY>，底層以幾何演算法自動解析，完全不需要 API Key。
• 軌道 2 (AI 語意抽取軌道)：只有在財報格式非標準、或員工人數深埋在英文散文中時，才需要交由 AI Agent 協助語意判定。
使用者如何知道「何時該用 AI？」以及「不同的 AI Agent 該如何使用？」
| AI 環境 / 工具 |
何時使用 (Trigger Conditions) |
如何使用 (Exact Instructions) |
| Antigravity / Coding Agent |
需要全自動下載 PDF、無損轉 Markdown、審計指標並重新編譯發布。 |
在對話中直接輸入：「請新增 [公司名稱]，下載歷年財報並加入儀表板」，Agent 會自動載入專屬 Skill 執行 7 步驟。 |
| Gemini 2.5 / 1.5 Pro (Google AI Studio) |
手邊有長達 300 頁的原生 10-K PDF，想快速抓出損益表與人均產值。 |
直接上傳 PDF，搭配本專案的 fininacial_prompt.md 提示詞，10 秒內獲取結構化 JSON。 |
| ChatGPT / Claude (網頁免費版) |
需要撰寫 16:9 董事會報告、投資戰略簡報、或分析同業競爭優劣勢。 |
在儀表板點擊「一鍵複製 Markdown (Copy Markdown)」或複製研究筆記，直接貼給 ChatGPT / Claude，讓其輸出專業投影片文字。 |
| Cursor / Windsurf / Copilot |
在本地 IDE 裡進行二次開發、擴充圖表或整合自訂內部 ERP 財務數據。 |
讓 Agent 讀取本專案的 AGENTS.md 與 SKILL.md，即可自動依循規範進行安全維護。 |