Baoyan Radar

Baoyan Radar Platform

Build a reusable 保研情报雷达 that can be published publicly while keeping each user's personal academic profile private.

Track both:

• official admissions notices from graduate-school / college websites
• community experience and evaluation pages such as Zhihu posts,经验贴,面经汇总, or lab-evaluation pages

Core rule

Do not hard-code the student's personal information into tracked files.

Keep public repository content limited to:

• generic monitoring code
• example config templates
• extraction and scoring logic
• local state schema

Keep private local content in untracked files such as:

• config/profile.local.json
• config/targets.local.json
• .env
• state/

Quick start

Fastest setup for a cloned repo:

Terminal wizard:

bashCopy code
python3 scripts/setup_clone.py

Local web UI:

bashCopy code
python3 scripts/setup_web.py

The setup flow will:

• collect the user's private profile locally, including detailed project and competition experience
• generate config/profile.local.json
• generate config/targets.local.json from built-in school presets after auto-positioning the user into matching tiers
• optionally install the daily scheduler for the current OS (macOS LaunchAgent or Windows Scheduled Task)
• optionally run one immediate test scan / push

Manual setup from examples is still supported:

bashCopy code
mkdir -p config state
cp references/profile.example.json config/profile.local.json
cp references/targets.example.json config/targets.local.json

Run one scan:

bashCopy code
python3 scripts/baoyan_radar.py once \
  --profile config/profile.local.json \
  --targets config/targets.local.json \
  --db state/radar.db

Run continuous monitoring:

bashCopy code
python3 scripts/baoyan_radar.py run \
  --profile config/profile.local.json \
  --targets config/targets.local.json \
  --db state/radar.db \
  --interval-min 60

Install a daily scheduled run with the cross-platform wrapper:

bashCopy code
python3 scripts/install_daily_schedule.py \
  --profile config/profile.local.json \
  --targets config/targets.local.json \
  --db state/radar.db \
  --hour 9 --minute 0 \
  --push-mode digest \
  --send-empty-digest

On macOS this creates a LaunchAgent. On Windows it creates a Scheduled Task.

Run the local dashboard after building the frontend:

bashCopy code
cd dashboard
npm install
npm run build
cd ..
python3 scripts/run_dashboard.py

Default address: http://127.0.0.1:8787/

Use the dashboard when the user needs a local platform view for filtering notices, tracking statuses, adding notes, seeing recent deadlines, or triggering scans interactively.

Analyze a single notice page:

bashCopy code
python3 scripts/baoyan_radar.py inspect \
  --url https://example.edu.cn/notice/123

Score a mentor or lab page against the student's profile:

bashCopy code
python3 scripts/baoyan_radar.py score \
  --url https://example.edu.cn/lab/pi

Generate a first-draft outreach email from local profile plus a mentor page:

bashCopy code
python3 scripts/baoyan_radar.py draft-email \
  --url https://example.edu.cn/lab/pi \
  --mentor-name 张老师

What the bundled script does

once

• load the private student profile and monitoring targets
• fetch target pages
• extract candidate links
• separately handle official and experience sources
• match admissions / experience keywords and profile keywords
• deduplicate by URL in SQLite
• summarize matched notices or experience pages
• optionally push matched summaries to Feishu webhook
• support either item push mode (one message per hit) or digest push mode (one daily summary)

run

Repeat once forever with a configurable interval.

inspect

Fetch one page and heuristically extract:

For official pages:

• school / college / project title
• absolute deadline text
• required materials
• assessment form
• profile-fit score

For experience pages:

• background / BG-like clues
• interview or machine-test notes
• suggestions and lab / school evaluation snippets
• profile-fit score

score

Fetch one mentor or lab page and compute a transparent keyword-overlap fit score using:

• target disciplines
• research keywords
• project keywords
• student project summaries

draft-email

Produce a structured, reusable 套磁信草稿 using the private profile plus the fetched page context.

Privacy model

If the user wants to publish the project on GitHub, do this by default:

1. commit only example configs from references/
2. add config/*.local.json, .env, and state/ to .gitignore
3. never commit real student name, rank, GPA, phone, email, Feishu webhook, or target list unless the user explicitly wants that

Community-source note:

• Prefer stable page URLs, collection pages, 专栏页, or manually curated post lists over fragile site-search result pages.
• Large platforms like Zhihu may change HTML or add anti-bot protections, so keep community sources configurable and expect occasional source-specific tuning.
• By default, experience sources are treated as single pages and do not follow page links. Set follow_links: true only for curated collection pages where link expansion is actually desired.

Tuning guidance

Prefer adding more precise target-page metadata before making the matcher more complex.

Useful tuning levers:

• source-specific include keywords
• source-specific exclude keywords
• profile research keywords
• project keywords and short project summaries
• minimum fit score threshold
• monitoring interval

References

Read these only when needed:

• references/profile.example.json for the private profile schema
• references/targets.example.json for target-page schema
• references/presets.cn-cs.json for built-in CS school/source presets used by the clone-setup wizard
• references/quickstart.md for clone-and-run onboarding
• references/privacy-and-layout.md for repository layout and Git hygiene
• references/scheduler.md for daily scheduling on macOS and Windows
• references/preset-expansion.md for how to keep expanding school / college / lab / mentor presets

Files written by the script

• SQLite DB path passed via --db
• optional local logs if the user redirects output or runs under a scheduler

Keep state files local and untracked unless the user explicitly wants to archive them.