Advisor Outreach

Advisor Outreach

Core Principle

Prioritize research fit over school prestige. Use 985/211/Double First Class status, discipline strength, and geography only after advisor-topic alignment and evidence quality are assessed.

Produce outreach that sounds deliberately written for the specific advisor. Avoid generic compliments, over-formal bureaucratic wording, inflated claims, and repeated template phrasing.

Workflow

1. Gather inputs:
   • User CV/resume, transcript, and optional personal statement.
   • User's outreach template or preferred tone.
   • Target advisor list, or a request to search for advisors.
   • User's research direction, degree target, school constraints, and geographic preferences.
2. Extract the user's reusable profile:
   • Education, GPA, rank, core courses, English scores, awards, competitions.
   • Research topics, methods, papers, patents, project responsibilities, tools, and evidence of independence.
   • A concise "advantage bank" that can be selectively reused per advisor.
   • For extraction details, read references/material-extraction.md.
3. Build or evaluate advisor profiles:
   • Advisor name, school, school type, school/discipline strength, lab or group, recent papers, projects, admissions hints, and reliable links.
   • For source and evidence rules, read references/advisor-profile.md.
4. Tier advisors:
   • A: highly aligned, active, and worth first contact.
   • B: broadly aligned and suitable as a main backup.
   • C: weakly aligned, exploratory, or strategic backup.
   • For scoring and ranking rules, read references/matching-rubric.md.
5. Generate outputs:
   • Advisor ranking table with evidence links and recommended contact priority.
   • One personalized email per advisor, not a mass-mail variant with only names swapped.
   • Optional subject lines, follow-up emails, and tracking table.
   • For format, read references/output-schema.md.

School-First Mode

When the user wants recommendations by school, group candidates by university first, then tier advisors inside each school. Do not stop after finding one or two excellent fits at a school; continue searching the same school for adjacent faculty in medical image analysis, computer vision, pattern recognition, multimodal learning, brain imaging, AI for healthcare, weak/semi-supervised learning, segmentation, or trustworthy/robust AI that can connect to the user's CV.

Use four tiers when the advisor pool is broad:

• S: direct overlap with the user's current research, such as weak/semi-supervised medical image segmentation, 3D vessel/brain image segmentation, multimodal brain imaging, or medical foundation models.
• A: strong adjacent fit, such as medical image analysis, AI-assisted diagnosis, multimodal medical data mining, or neuroimaging AI.
• B: transferable CV/ML fit, such as computer vision, segmentation, representation learning, multimodal learning, graph learning, or robust learning without clear medical focus.
• C: strategic backup, such as broad AI, data mining, pattern recognition, or biomedical engineering faculty where the fit is plausible but needs a careful email angle.

For school-first searches and output, read references/school-first-mode.md.

Email Generation Rules

Use the user's template as a style reference, not as fixed text. Preserve accurate facts from the user's materials, but rewrite the email around the advisor's actual work.

Each email should include:

1. Subject line with year, purpose, name, school, and concise identity.
2. Opening that is polite but not stiff.
3. A compact self-introduction using only the strongest relevant facts.
4. A research-experience paragraph that emphasizes concrete work done by the user.
5. A specific advisor-fit paragraph grounded in the advisor's papers, lab direction, or projects.
6. A forward-looking paragraph explaining what the user wants to continue learning in that group.
7. A concise close asking whether there is an opportunity to communicate further.

Do not claim the user read a paper, understands a project deeply, or has a strong preference for an advisor unless the source evidence supports it. If evidence is weak, use careful wording such as "我注意到您团队近年来关注..." rather than "我认真研读了...".

For reusable sentence patterns and anti-template guidance, read references/email-style.md.

Batch Mode

When the user provides multiple advisors, create one table plus individual email drafts. Keep shared achievements consistent, but vary:

• The research-fit paragraph.
• Which user experiences are emphasized.
• The proposed future direction.
• The subject line if the advisor's field differs.

Flag missing advisor evidence instead of inventing specifics. If only names and schools are provided, first ask for permission to search or request advisor homepage/paper links.