Johns Hopkins academic project

AI Inbox Triage & Drafting Assistant

A human-reviewed system for turning a crowded inbox into prioritized work, grounded summaries, and safer first-response drafts.

Scope: midterm project for applied generative AI study at Johns Hopkins, submitted under the assignment title AI Email Secretary. It uses a synthetic business scenario and sample email dataset; it is not connected to a live inbox and does not send messages autonomously.

Problem

A manager facing a mixed inbox needs to separate genuine emergencies and same-day deadlines from routine updates, informational messages, and spam. A useful assistant must reduce review time without mistaking urgent language for actual urgency—or making commitments on the manager’s behalf.

Approach

  1. Filter deterministically. Apply the Yesterbox method with pandas so only the previous day’s messages enter the workflow.
  2. Classify with guardrails. Sort messages into six defined categories while checking the full body, recipient, context, and actual deadline.
  3. Generate one item at a time. Produce source-grounded summaries, next steps, and first-response drafts for each critical email.
  4. Protect consequential decisions. Prevent the model from inventing approvals, commitments, attendance, attachment reviews, or unsupported facts.
  5. Evaluate against the source. Use structured LLM-as-a-judge scoring for relevance, clarity, and actionability while preserving traceability to each email.

Result

The workflow processed all 51 previous-day messages in the sample dataset. It identified 23 critical messages, generated complete summaries for 13 urgent and 10 deadline-driven emails, produced 23 first-response drafts, and evaluated all 23 drafts against their source messages.

The lower actionability score exposed the project’s most useful next step: make safe follow-up timing, ownership, and requests for missing information more specific without allowing the model to fabricate a decision.

What it demonstrates

The project combines deterministic data processing, constrained prompt design, one-to-one completeness checks, structured evaluation, and explicit human oversight. Its conclusion is intentionally modest: the system can support inbox triage and drafting, but a person must verify classifications, facts, dates, and commitments before any response is sent.

← Back to AI & Media Operations