Medical Researcher / Principal Investigator
Profile. PI, clinical investigator, post-doc, or research-track physician. 200+ messages/day across journal alerts, NIH/sponsor communications, IRB notices, society newsletters, trainee updates, manuscript reviews, and coordinator threads. HIPAA-aware by institutional default.
The two-hour daily problem
90 minutes of newsletter / journal-alert skimming first thing, another 30 of inbox triage between clinic and bench. Most of it is mechanical — categorize, file, skim-and-discard. None of it is research.
The outcome — the newsletter digest is the killer feature
All 30+ subscription feeds collapse into one 7 AM digest, 80–120 article cards organized by source. An estimated 60–90 minutes/day recovered for a subscription-heavy inbox. Scan it with your first coffee, click through to the 5–8 articles worth reading, archive the rest by closing the tab.
Plus the rest of the stack
- Sponsor mail labeled
sponsor, with a drafted reply waiting for review. - IRB notices auto-filed into a dedicated folder.
- Trainee questions routed to the daily catch-up window.
- An end-of-day triage-activity digest — no "what is the AI doing while I'm not looking" anxiety.
- HIPAA mode on the patient-correspondence account for institutional review.
Two v0.2.0 features built for this persona
A Regulatory Deadline Clock watches for the mail that starts an FDA IND / EU CTR SUSAR / IRB reporting clock (detection is opt-in), tracks it on a working-day calendar, and escalates before it lapses — a human confirms every clock and the AI never computes the date. A Topic / Study Digest pulls every message tied to one study — sponsor, IRB, and coordinator inboxes at once — into a single scheduled digest, with a thumbs up/down that teaches which senders matter. PHI never leaves your servers.
The Deadlines view — an armed IRB continuing-review clock (Running, working-day countdown, explicit due date, Acknowledge / Mark-submitted / Dismiss). Detection is opt-in and a human confirms every clock; sponsor FDA/EU deadlines surface as informational only.
Total recovered: an estimated 1.5–2 hours/day, with full transparency over what the system handled and what's waiting for you.
Law Firm / Legal Department (Privileged Inbox)
Context. A litigation boutique, in-house legal team, or solo practitioner whose inbox is a running record of privileged client communications and attorney work-product. The volume is real — client threads, opposing counsel, court notices, e-filing alerts — but the sharper concern is where the mail goes the moment an AI reads it.
The problem with routing privileged mail through cloud AI
Send a privileged inbox through a hosted, third-party AI service and you have handed a client's confidential communications to an outside party that now received them. That outside party is a new custodian a subpoena can reach in discovery, and its receipt of the mail is a disclosure the other side can point to when arguing the privilege was waived. The convenience of a hosted model quietly widens the circle of who holds the file.
The outcome — the AI stays inside the firm's walls
The classifier runs on the firm's own hardware; privileged mail and work-product are read and sorted on-box and never transit an outside model. There is no third-party AI vendor to name in a subpoena and no external disclosure to build a waiver argument from — the mail stays under the firm's sole control, the way the file room already does. Sensitive-matter routing, drafted replies held for a lawyer's review (never auto-sent), and a tamper-evident log of who opened which matter round it out.
This is an infrastructure-and-strategy posture, not legal advice — confirm the privilege and discovery implications for your jurisdiction with your own counsel.
Solo Practitioner Clinic (1–5 person practice)
Context. M365 or Google Workspace-based. HIPAA-regulated by default. Need to demonstrate compliance posture to insurers + auditors.
Outcome
- HIPAA-mode accounts with PHI-scrubbed logs, and an audit chain on every staff access.
- Drafted patient-correspondence replies held for clinician review (never auto-sent).
- Delegate access for front-desk staff, scoped to specific mailboxes with audit-row stamping — sensitive operations stay with the owner.
- A paginated compliance view over the tamper-evident audit log for periodic review.
Fractional CTO / Consulting Firm
Context. Multiple client mailboxes (separately authenticated). Need to keep client communications isolated for confidentiality. Calendar coordination across 8–12 active clients per week.
Outcome
- Per-client account with isolated rules and routes.
- A per-client writing-style profile, so replies sound right for each engagement.
- Meeting-request intercept on every account, with the right working-hours config per client.
- Watch rules escalate "fire-drill" client emails to SMS instantly.
This is how I run my own consulting practice.
Researcher Cohort Tooling
Context. A lab with N researchers all using shared mail infrastructure. Each researcher has their own writing style and their own preferences. Privacy: one researcher shouldn't see another's mail or drafts.
Outcome
- Multi-user install with role-based access — each researcher gets their own dashboard, categories, and digest.
- The lab PI has delegate access to specific researcher mailboxes (with audit-row stamping).
- One Ollama host serves the whole lab: one container, one TLS cert, one backup.
Privacy-First Homelab Operator
Context. A self-hosting enthusiast who wants to run everything on their own machines. (For the technically-inclined: containers on Podman, reached over a Tailscale mesh, with the AI model already running locally on their own GPU via Ollama.)
Outcome
- Install + configure independently from the public repo at github.com/Unlimited-Data-Works-LLC/Email-Triage.
- Pull the cosign-signed image from
ghcr.io/unlimited-data-works-llc/email-triage, verify the SLSA-3 provenance, run on Podman with a systemd quadlet. - Air-gap installable via the hash-pinned embedding-runtime tarball — same verification path as the auto-download.
- Audit chain and HIPAA-mode features work exactly the same as in any other deployment — the privacy posture isn't gated behind an engagement.