🟢 #4 feat(extraction): Phase 3.5a - MAR/Vitals lenses + DocView memory-leak fix +1,740 -178 2mo ago self by
rmerk ↗ analysis ↗ GitHub ## Summary
rmerk/mike
rmerk is bending Mike into a Minnesota medical-malpractice records engine that mines charts and builds litigation-ready chronologies.
This fork, from rmerk, has a clear and unusually specific ambition: turn Mike into a working platform for Minnesota medical-malpractice cases. Rather than staying a general document tool, it's being built out in phases - scaffold a malpractice matter, mine the medical records, and hand the user back a citation-linked chronology, with each piece wired to specific Minnesota statutes and case law.
What you'd encounter is a records-analysis workflow aimed at litigators: extraction that pulls structured events out of dense medical PDFs, purpose-built lenses for medication administration (MAR) and vitals, and a fast timeline view that renders the whole event log as a sortable, source-linked chronology. There's real attention to correctness under legal pressure - including a compliance gate that catches peer-review material hiding in scanned pages a text-only pass would miss.
Under the hood, rmerk has also broadened Mike's AI options, adding NVIDIA-hosted models and making a vision-capable model the default extraction brain. If you want to see how far the med-mal buildout has gone, the fork is worth a click through to GitHub.
What's in it
- Minnesota med-mal platform A phased buildout that scaffolds a malpractice matter, mines the records, and produces a chronology - grounded in Minnesota statutes and case law.
- Medical record extraction An extraction pipeline that turns dense, often-scanned medical PDFs into a structured, citable event log.
- MAR & Vitals lenses Purpose-built views over the event log for medication administration and vitals - the details that matter most in a malpractice review.
- Chronology timeline A fast, sortable timeline that renders the extracted events as a citation-linked chronology, built to load quickly without an extra model call.
- Peer-review compliance gate A vision prescan that flags peer-review material appearing only in a page's image layer, halting before those pages slip through unreviewed.
- Broader model support Adds NVIDIA-hosted models to Mike and makes a vision-capable model the default engine for reading records.
Direction
summarizationcompliancediscovery
Activity
✅ #3 feat(timeline): med-mal chronology view backed by event log +485 2mo ago self by
rmerk ↗ analysis ↗ GitHub ## Summary
✅ #2 feat(extraction): Phase 2 med-mal extraction pipeline +6,900 -21 2mo ago self by
rmerk ↗ analysis ↗ GitHub ## Summary
A five-phase buildout that scaffolds a malpractice case, mines the medical records, and hands you a chronology - every piece wired to specific Minnesota statutes and case law.
📝 rmerk bolts NVIDIA onto Mike and quietly changes the default brain 1 commit 2mo ago infrastructureintegration draft
A fourth AI provider lands, and fresh installs now think with NVIDIA-hosted models instead of Google's Gemini.
✅ #1 docs: add CLAUDE.md with architecture orientation +82 2mo ago self by
rmerk ↗ analysis ↗ GitHub Adds top-level CLAUDE.md so Claude Code sessions get a fast architectural orientation. Defers user setup to README.md to avoid duplication.
## Summary
## Summary
📝 Add a CLAUDE.md orientation file for coding agents 1 commit 2mo ago minor change
📝 Add a CLAUDE.md to orient coding agents to the repo 0 commits — minor change
📝 Phase 3.5a: MAR and Vitals timeline lenses, Kimi K2.6 as extraction default, DocView memory-leak fix 0 commits — discoverysummarization draft
rmerk's Phase 3.5a PR adds MAR and Vitals views over the existing `document_events` log, switches the extraction model to NVIDIA Catalog's Kimi K2.6 (vision), and addresses a memory leak in the extraction page that cause…
📝 Add CLAUDE.md for fast Claude Code orientation 0 commits — minor change
📝 Phase 2 extraction pipeline: §145.64 vision prescan closes compliance gap on scanned pages 0 commits — compliancediscovery draft
rmerk's Phase 2 med-mal extraction adds a peer-review compliance gate that the earlier text-only prescan left open. On scanned pages where peer-review marker phrases appear only in the raster - not the text layer - the o…
📝 Phase 3 chronology view: Timeline renders event log with bbox sync, no backend changes 0 commits — summarizationdiscovery draft
rmerk's Phase 3 adds a Timeline view at `/projects/[id]/timeline/[docId]` that renders the Phase 2 event log as a sortable, citation-linked chronology. The roadmap set a hard gate - under two seconds, no LLM call - and t…
Threads of work (detailed view)
rmerk is turning Mike into a Minnesota med-mal machine
A five-phase buildout that scaffolds a malpractice case, mines the medical records, and hands you a chronology - every piece wired to specific Minnesota statutes and case law.
Minnesota med-mal records platform: five phases from research validation to chronology view
rmerk/mike spent a single day on May 11, 2026 shipping a vertical slice that turns Mike into a Minnesota-jurisdiction-aware med-mal records pipeline. The work spans research validation, project templates, a ~6,900-LOC extraction pipeline, and a citation-anchored chronology view - all grounded in specific MN statutes and Supreme Court authority.
rmerk bolts NVIDIA onto Mike and quietly changes the default brain
A fourth AI provider lands, and fresh installs now think with NVIDIA-hosted models instead of Google's Gemini.
Fourth LLM provider: NVIDIA API Catalog with Kimi K2.6 as new default
rmerk adds `backend/src/lib/llm/nvidia.ts` as a fourth provider alongside Claude, Gemini, and OpenAI, targeting NVIDIA's OpenAI-compatible endpoint at `integrate.api.nvidia.com/v1`. Kimi K2.6 becomes the new default model for chat, titles, and tabular review - replacing the prior Gemini default across the board.
Phase 3 chronology view: Timeline renders event log with bbox sync, no backend changes
rmerk's Phase 3 adds a Timeline view at `/projects/[id]/timeline/[docId]` that renders the Phase 2 event log as a sortable, citation-linked chronology. The roadmap set a hard gate - under two seconds, no LLM call - and this satisfies it by reading an existing endpoint with no new backend code or migrations.
Phase 2 extraction pipeline: §145.64 vision prescan closes compliance gap on scanned pages
rmerk's Phase 2 med-mal extraction adds a peer-review compliance gate that the earlier text-only prescan left open. On scanned pages where peer-review marker phrases appear only in the raster - not the text layer - the original check would have passed them through to event extraction. The new path rasterizes those pages, asks Claude whether markers are visible, and halts before any event row is written.
Phase 3.5a: MAR and Vitals timeline lenses, Kimi K2.6 as extraction default, DocView memory-leak fix
rmerk's Phase 3.5a PR adds MAR and Vitals views over the existing `document_events` log, switches the extraction model to NVIDIA Catalog's Kimi K2.6 (vision), and addresses a memory leak in the extraction page that caused RSS to grow without bound on large Epic PDFs.
Pull requests (detailed view)
🟢 Open (1)
by
rmerk · opened 2mo ago ✅ Merged (3)
by
rmerk · opened 2mo ago · merged 2mo ago by rmerk by
rmerk · opened 2mo ago · merged 2mo ago by rmerk by
rmerk · opened 2mo ago · merged 2mo ago by rmerk ⛔ Closed without merge (2)
by
rmerk · opened 2mo ago · closed 2mo ago by
rmerk · opened 2mo ago · closed 2mo ago