Mike Swanson
46485af009
Author: Mike Swanson Machine: DESKTOP-0O8A1RL Timestamp: 2026-05-12 17:13:53
178 lines
6.5 KiB
Markdown
178 lines
6.5 KiB
Markdown
# Grabb Durando — AI Demand Review System
|
||
|
||
**Client:** Grabb Durando (plaintiff personal injury law firm, Arizona)
|
||
**Contact:** Robert Grabb — rgrabb@grabblaw.com
|
||
**Project start:** 2026-05-12
|
||
**Status:** Scoping / pre-quote
|
||
|
||
---
|
||
|
||
## What They Want
|
||
|
||
Robert has designed a full AI-assisted workflow for preparing pre-suit demand packages (Phase Two Package, delivered 2026-05-11). He wants ACG to design and implement the technical system.
|
||
|
||
The workflow:
|
||
1. Staff copies a 11-subfolder template into a client case file
|
||
2. Drops relevant documents into categorized folders (medical records, police report, bills, liens, etc.)
|
||
3. Uploads to a "firm-approved AI" — runs a standard prompt
|
||
4. AI produces: case snapshot, liability summary, medical chronology, specials summary, missing items, strengths/weaknesses, draft demand letter
|
||
5. Staff saves AI output to folder 11 (AI Output)
|
||
6. Attorney reviews and approves before anything is sent
|
||
|
||
The "firm-approved AI" is the piece ACG is building.
|
||
|
||
---
|
||
|
||
## Their Environment
|
||
|
||
- **Case management:** Leap (legal case management software)
|
||
- **File storage:** Network/file server (not SharePoint/OneDrive)
|
||
- **M365:** Yes (existing MSP client, ACG-managed tenant)
|
||
- **Inky/GuruProtect:** Yes (visible in email headers)
|
||
|
||
---
|
||
|
||
## Implementation Decision
|
||
|
||
**Custom Claude API application** — a web UI where staff:
|
||
1. Selects a case, uploads documents from each category
|
||
2. Sees a pre-flight checklist (which categories have docs, which are empty)
|
||
3. Clicks "Generate" — app sends documents + standard prompt to Claude API
|
||
4. Reviews structured output on screen
|
||
5. Downloads/saves output as DOCX or PDF to the AI Output folder
|
||
|
||
This approach:
|
||
- Gives the firm a single, controlled access point (no staff wandering to personal ChatGPT)
|
||
- Enforces the standard prompt (staff can't accidentally use the wrong one)
|
||
- Creates an audit log (who ran what case, when)
|
||
- Is hosted and managed by ACG
|
||
|
||
---
|
||
|
||
## Standard AI Prompt (from Robert's docs)
|
||
|
||
```
|
||
You are assisting a plaintiff personal injury law firm in Arizona with a pre-suit demand package.
|
||
|
||
Review the uploaded materials for this case. Prepare the following:
|
||
1. One-page case snapshot
|
||
2. Liability summary
|
||
3. Medical chronology by date and provider
|
||
4. Medical specials summary by provider
|
||
5. Summary of liens, LOPs, MedPay, or reimbursement issues if shown in the documents
|
||
6. Missing records, missing bills, or missing information
|
||
7. Case strengths
|
||
8. Case weaknesses or defense arguments
|
||
9. Questions for staff or attorney review
|
||
10. Draft settlement demand letter
|
||
|
||
Important instructions:
|
||
Do not invent facts. Do not assume facts not shown in the documents.
|
||
If something is unclear, say "unclear from the materials provided."
|
||
If documents are missing, identify what is missing.
|
||
Separate confirmed facts from assumptions.
|
||
Use a professional plaintiff personal injury tone.
|
||
Do not include legal citations unless specifically requested.
|
||
Do not send anything. This is a draft for attorney review.
|
||
|
||
After preparing the demand letter, create a short checklist of facts that staff or attorney must verify before the demand is sent...
|
||
```
|
||
|
||
---
|
||
|
||
## Document Categories (the 11-folder structure)
|
||
|
||
| # | Category |
|
||
|---|----------|
|
||
| 01 | Intake and Client Facts |
|
||
| 02 | Liability Documents |
|
||
| 03 | Photos and Videos |
|
||
| 04 | Insurance and Claim Info |
|
||
| 05 | Medical Records |
|
||
| 06 | Medical Bills |
|
||
| 07 | Liens and LOPs |
|
||
| 08 | Wage Loss |
|
||
| 09 | Adjuster Correspondence |
|
||
| 10 | Prior Offers and Demands |
|
||
| 11 | AI Output (where output is saved) |
|
||
|
||
---
|
||
|
||
## Work Breakdown & Quote Basis
|
||
|
||
### Phase 1 — Discovery & Design (4–6 hrs)
|
||
- Review Leap document structure and how docs are accessed/exported
|
||
- Define user roles (staff vs. attorney — read-only output?)
|
||
- Document size and format limits (PDF, DOCX, JPG, PNG, TXT)
|
||
- Decide hosting location (ACG-managed server or small VM)
|
||
- Design the app flow and UI wireframe
|
||
- Select Claude model (Sonnet for quality vs. Haiku for cost)
|
||
- Technical design doc
|
||
|
||
### Phase 2 — Development (18–28 hrs)
|
||
- Document upload UI with per-category slots and pre-flight checklist
|
||
- Document text extraction: PDF (pdfplumber/PyMuPDF), DOCX (python-docx), images (Claude vision)
|
||
- Claude API integration with the standard prompt
|
||
- Structured output display: each of the 10 sections rendered clearly
|
||
- Download output as DOCX or PDF
|
||
- Per-case audit log (who ran, when, which documents, which Claude model/version)
|
||
- User authentication (simple — not AD-integrated in v1)
|
||
- Basic admin panel: manage users, view audit log
|
||
|
||
### Phase 3 — Testing & Prompt Refinement (4–6 hrs)
|
||
- Test with sample/redacted case documents
|
||
- Tune prompt for output quality
|
||
- Review with Robert and one staff member
|
||
- Edge cases: missing docs, large PDFs, image-only docs
|
||
|
||
### Phase 4 — Deployment (3–4 hrs)
|
||
- Deploy to ACG-managed server (Linux VM or Windows)
|
||
- SSL certificate
|
||
- Configure for firm's network access (internal or VPN-accessible)
|
||
- Backup configuration
|
||
|
||
### Phase 5 — Training & Documentation (3–4 hrs)
|
||
- 1–2 hour staff training session
|
||
- Written user guide (PDF)
|
||
- Admin documentation for ACG
|
||
|
||
**Total estimated hours: 32–48**
|
||
|
||
### Ongoing / Recurring
|
||
- Claude API usage: ~$0.05–$0.30 per demand package (Sonnet, depends on doc volume) — pass-through or light markup
|
||
- Monthly hosting/maintenance: included in MSP contract or separate line item (~$50–75/month)
|
||
- Annual prompt review/update: 1–2 hrs
|
||
|
||
---
|
||
|
||
## Quote Ranges
|
||
|
||
| Tier | Hours | Rate | Range |
|
||
|------|-------|------|-------|
|
||
| Low | 32 hrs | $125/hr | $4,000 |
|
||
| Mid | 40 hrs | $135/hr | $5,400 |
|
||
| High | 48 hrs | $145/hr | $6,960 |
|
||
|
||
Recommend quoting as a flat-fee project with a defined scope, plus a change-order clause for Leap integration if they want direct file server/Leap connectivity in v2.
|
||
|
||
---
|
||
|
||
## Open Questions (for discovery call)
|
||
|
||
1. How many users will need access? (affects auth complexity)
|
||
2. Does Leap have an API or document export format? (affects file server integration)
|
||
3. Where are case documents stored on the file server — is there a predictable folder structure per case?
|
||
4. Does Robert want staff to manually upload each document, or pull from a case folder path automatically?
|
||
5. Any requirement to retain AI output within Leap (vs. just saving to the file server)?
|
||
6. Timeline expectation?
|
||
7. Budget ballpark?
|
||
|
||
---
|
||
|
||
## Source Materials
|
||
|
||
Stored in: `clients/grabb-durando/extracted-attachments/`
|
||
- `AI_Demand_Review_Phase_Two_Package.docx` — full Phase Two package
|
||
- `AI_Demand_Review_Checklist (2).docx` — standalone checklist
|
||
- `zip-contents/AI_Demand_Review_Phase_Two/AI Demand Review - TEMPLATE/` — full folder template + all source files
|