Bob the Bot

Portfolio Announcements Intelligence Agent

Bob the Bot is an automated AI agent that monitors ASX announcements for companies in my investment Portfolio, analyses important updates, and delivers a daily intelligence briefing.

The goal is simple:

Read everything so I don't have to. Surface the signal and ignore the noise.

Instead of manually checking ASX announcements every day, Bob retrieves new filings, downloads important documents, runs AI analysis when appropriate, and sends a concise summary.

System Overview ┌──────────────────────────┐ │ My Portfolio │ │ tickers.yaml │ └─────────────┬────────────┘ │ ▼ ┌──────────────────────────┐ │ ASX Announcements │ │ Data Retrieval │ └─────────────┬────────────┘ │ ▼ ┌──────────────────────────┐ │ Bob the Bot │ │ agent.py Engine │ └─────────────┬────────────┘ │ ┌────────────────┼────────────────┐ ▼ ▼ ▼

┌───────────────┐ ┌────────────────┐ ┌────────────────┐ │ Announcement │ │ Document │ │ AI Analysis │ │ Classification │ │ Retrieval │ │ (OpenAI API) │ └───────┬───────┘ └───────┬────────┘ └────────┬───────┘ │ │ │ ▼ ▼ ▼

┌───────────────┐ ┌────────────────┐ ┌────────────────┐ │ Portfolio │ │ Google Drive │ │ Strawman Post │ │ Intelligence │ │ Document Store │ │ Draft Generator│ └───────┬───────┘ └────────┬───────┘ └────────┬───────┘ │ │ │ └──────────────┬────┴─────────────┬─────┘ ▼ ▼

                ┌───────────────┐
                │  Daily Email  │
                │ Intelligence  │
                │    Digest     │
                └───────────────┘

What Bob Does

Every day Bob automatically:

Monitors ASX announcements for companies in my Portfolio

Identifies announcements released within the last 24 hours

Avoids reporting the same announcement on consecutive days by tracking seen announcement IDs between runs

Filters out routine filings and low-impact updates

Downloads PDFs for meaningful announcements

Extracts text from reports and presentations

Runs AI analysis on major announcements

Generates short summaries for quick review

Saves important documents to Google Drive

Sends a daily intelligence briefing email

Bob focuses on signal over noise.

Example Output

Bob sends a daily briefing structured like this:

Bob the Bot Daily Announcements Digest — last 24 hours

HIGH IMPACT

DRO — FY Results

Revenue growth strong but margin compression continues.

Analysis: [Detailed AI analysis]

Open: https://www.asx.com.au/...

Drive: https://drive.google.com/...

---

MATERIAL

TWE: Retirement of Chief Financial Officer So what: Leadership transition likely operational rather than strategic.

Open: https://www.asx.com.au/...

---

FYI

ABB: Appendix 3Y Director Interest Notice So what: Routine filing. Portfolio Monitoring

Companies are defined in:

tickers.yaml

Example:

asx:

• DRO
• AR9
• CAT
• ABB
• TWE

Bob retrieves announcements for each ticker.

Announcement Source

Bob retrieves announcements directly from the ASX announcements feed.

Example endpoint:

https://www.asx.com.au/asx/v2/statistics/announcements.do?asxCode=DRO

Information extracted includes:

announcement title

release date

announcement URL

PDF document link

Only announcements from the last 24 hours are processed.

Announcement Classification

Announcements are automatically categorised:

Category Description RESULTS_HY_FY Half year or full year results ACQUISITION M&A transactions CAPITAL_OR_DEBT_RAISE Equity issuance or refinancing CONTRACT_MATERIAL Major contracts or strategic updates OTHER Routine filings

Classification uses keyword detection and content signals.

Document Retrieval

For important announcements Bob retrieves the PDF.

Retrieval pipeline:

Step 1 — Direct Download

Using Python requests.

Step 2 — Browser Simulation

If blocked by the ASX consent page:

Playwright

simulates a browser to retrieve the file.

Step 3 — HTML Fallback

If the PDF cannot be retrieved, Bob extracts text from the webpage.

Text Extraction

When a PDF is downloaded:

pypdf

extracts the text.

Bob also checks that the content is meaningful to avoid analysing:

empty documents

ASX legal disclaimer pages

broken downloads

AI Analysis

Bob uses the OpenAI API.

Default model:

gpt-4o-mini

AI analysis is used selectively to control cost.

Deep Analysis Triggers

Detailed AI analysis runs only for major events:

Half-year results

Full-year results

Acquisitions

Capital raises

Debt refinancings

These produce:

• full AI analysis • a Strawman-ready investor post • saved documents in Google Drive

Quick Summaries

Most announcements receive a short two-line summary:

DRO: $21.7m Western Military Contracts So what: Adds to defence pipeline but unlikely to materially move revenue this year. Google Drive Storage

Important documents are automatically uploaded to Google Drive.

Stored documents include:

results reports

investor presentations

acquisition documents

capital raise materials

Drive links are included in the email.

Email Alerts

The daily digest contains:

HIGH IMPACT

Major announcements requiring deep analysis.

MATERIAL

Price-sensitive announcements requiring attention.

FYI

Routine announcements included for completeness.

SILENCE

If no announcements occurred:

No announcements in the last 24 hours.

When this happens, Bob now also includes:

• A Joke of the Day
• A Political Cartoon of the Day link Special Routing

Some tickers trigger additional alerts.

Example:

AR9

If AR9 appears in the announcements:

Bob sends a separate email digest to my brother.

Architecture

Core system files:

agent.py

Main orchestration engine.

playwright_fetch.py

Browser automation for retrieving ASX PDFs.

prompts.py

AI prompts used for announcement analysis.

tickers.yaml

Portfolio company list.

Workflow

Bob runs automatically using GitHub Actions.

Example schedule:

5 1 * * *

Which corresponds to:

09:05 Singapore Time

Bob runs entirely in the GitHub cloud environment.

Your computer does not need to be on.

Required Environment Variables

Configured using GitHub Secrets:

EMAIL_FROM EMAIL_TO EMAIL_APP_PASSWORD OPENAI_API_KEY GDRIVE_SERVICE_ACCOUNT_JSON GDRIVE_FOLDER_ID BROTHER_EMAIL Forcing a Re-run for a Specific Ticker

Bob tracks every announcement it has already processed in state_seen.json. Seen entries expire after 72 hours, but within that window any announcement that was processed in a previous run is skipped — even if you trigger Bob manually on the same day.

To re-process announcements for one or more tickers without clearing the entire state file:

1. Go to the Actions tab in GitHub.
2. Select the Daily Announcement Digest workflow.
3. Click Run workflow.
4. Enter the ticker(s) in the "Comma-separated ASX tickers to force re-run" field (e.g. NHC or NHC,BHP).
5. Click Run workflow.

Bob will bypass the dedup check only for those tickers and re-process any of their announcements that fall within the current HOURS_BACK window — all other tickers remain deduplicated as normal.

This can also be triggered by setting the FORCE_RERUN_TICKERS environment variable when running locally:

FORCE_RERUN_TICKERS=NHC python agent.py

Safety Limits

To prevent runaway API usage:

MAX_PDFS_PER_RUN = 10 MAX_LLM_CALLS_PER_RUN = 15 MAX_ANNOUNCEMENTS_PER_TICKER = 12 Technology Stack

Bob is written in Python.

Key libraries:

requests beautifulsoup4 pypdf playwright openai google-api-python-client pyyaml Deployment

Bob runs fully in the cloud via:

GitHub Actions

Execution environment:

Python 3.11 GitHub Hosted Runner Future Improvements

Potential upgrades:

LSE announcement monitoring

broker research ingestion

earnings model extraction

insider trading detection

Slack or Telegram alerts

portfolio dashboard visualisation

Why Bob Exists

The ASX produces hundreds of announcements every week.

Most are irrelevant.

Bob's job is to read everything so I don't have to, highlight the important information, and deliver a clean intelligence briefing for my Portfolio.

---

Wally (Watchlist 52-week low agent)

Wally is a second, separate agent that screens multiple watchlists for stocks trading near their 52-week lows.

What Wally does (V1)

• Loads watchlists from watchlists/*.yaml
• Fetches latest price + 52-week low/high per ticker
• Flags stocks within 5% of 52-week low:
  • distance_to_low_pct = ((current - low_52w) / low_52w) * 100
  • flagged (near_low) when distance_to_low_pct <= 5
• Also flags stocks at or below a configured per-ticker target ("buy") price:
  • below_target when current <= target_price
  • distance_to_target_pct = ((current - target_price) / target_price) * 100
  • a ticker is flagged when near_low OR below_target, and below-target tickers get the same treatment as near-low ones (range + value charts, email detail, dashboard row)
• Creates:
  • 10-year price-vs-value chart when valuation config exists
• Sends an HTML email report (with chart attachments)
• Writes machine-readable JSON output under outputs/YYYY-MM-DD/

Watchlist files

Supported format — a plain string list still works exactly as before:

name: JM Watch List
tickers:
  - ARB.AX
  - QOR.AX

Target ("buy") prices are optional and fully backwards-compatible. Any of the three forms below may be mixed in one file; plain strings simply carry no target:

name: JM Watch List
tickers:
  - ARB.AX                 # plain string — no target
  - ticker: CGS.AX
    target_price: 1.20     # per-entry mapping
  - ticker: QOR.AX
    buy_price: 3.50        # `buy_price` is an accepted alias (used by the TII list)

# ...or a separate top-level block keyed by ticker:
targets:
  ARB.AX: 30.00
  CGS.AX: 1.20

Loader rules: ticker keys are normalised with .strip().upper(); prices are coerced to float and anything <= 0 or non-numeric is ignored, so a bad value never produces a false flag. GBX pence prices may be written the way the source spreadsheet holds them ("500.00p", "9,500.00p") — the loader strips the thousands comma and trailing p to a numeric pence value (compare against a pence-quoted quote). When CGS.AX has target_price: 1.20, it is flagged the moment it trades at or below 1.20 and appears in the email with Target 1.20 and Trigger Below target (or Near 52w low + Below target when both apply).

Files used:

• watchlists/tii_watchlist.yaml
• watchlists/jm_watchlist.yaml
• watchlists/aussie_tech_watchlist.yaml
• watchlists/tii75_watchlist.yaml

Valuation config

Place per-ticker config at:

• valuations/<ticker_lower_with_underscores>.yaml
  • e.g. valuations/bhp_ax.yaml for BHP.AX

Wally uses configured EPS/dividend series to plot value lines. If missing, the report states: No valuation config found yet for this ticker.

Scheduling

Workflow: .github/workflows/wally_watchlists.yml

• Tue + Fri: standard watchlists (tii, jm, aussie_tech)
• Fri only: tii75 is attempted, then Python applies fortnightly gate
  • gate logic in wally.config.should_run_tii75
  • adjustable via env TII75_ANCHOR_ISO_WEEK

Combined email mode

As of the latest update, Wally supports combined email mode to consolidate all watchlist results into a single email:

# Run all watchlists (standard + TII75) with combined email
python -m wally.main --all-combined

# Run standard watchlists only with combined email
python -m wally.main --all-standard-watchlists --combined-email

# Force TII75 inclusion even if gated
python -m wally.main --all-combined --force

The GitHub Actions workflow now uses combined email mode by default. See docs/WALLY_AND_MASTER_ENGINE_CHANGES.md for details.

Required secrets / env

Wally reuses Bob email settings where possible:

• EMAIL_FROM (or EMAIL_USER)
• EMAIL_TO
• EMAIL_APP_PASSWORD (or SMTP_PASS)
• optional SMTP overrides: SMTP_HOST, SMTP_PORT, SMTP_USER

Local run commands

python -m wally.main --watchlist watchlists/jm_watchlist.yaml
python -m wally.main --all-standard-watchlists
python -m wally.main --watchlist watchlists/tii75_watchlist.yaml --force
python -m wally.main --tii75 --force

Sunday Sally (Weekly Valuation Review)

A new weekly agent is available under sunday-sally/ to review names near 52-week highs and assess valuation stretch vs history. See sunday-sally/README.md and .github/workflows/sunday_sally_weekly_review.yml.

Master Engine Alert (Super Investor)

The Master Engine Alert aggregates events from all agents (Bob, Wally, Ned) into a unified investor briefing with prioritized alerts.

How to run

Via GitHub Actions:

1. Go to the Actions tab in GitHub
2. Select "Master Engine Alert (Super Investor)" workflow
3. Click "Run workflow"
4. Configure options (include TII75, skip agents, etc.)
5. Click "Run workflow" button

Locally:

python run_master_investor.py
python run_master_investor.py --wally-tii75  # Include TII75 watchlist
python run_master_investor.py --no-ned --no-bob  # Skip specific agents
python run_master_investor.py --no-email  # Generate digest without sending email

Output locations

• Email: Sent to EMAIL_TO with subject "Johnny Master Investor Alert — {date} ({N} alert(s))"
• Files: Saved in outputs/YYYY-MM-DD/:
  • master_investor_digest.html - Full HTML digest
  • master_investor_digest.md - Markdown summary
  • master_investor_events.json - JSON archive

Scheduled runs

The workflow runs automatically daily at 00:00 UTC, after all individual agent workflows have completed.

See docs/WALLY_AND_MASTER_ENGINE_CHANGES.md for detailed documentation.