Grigorii Moskalev
Legal counsel and legal engineer.
11 years practicing law, 3 years writing Python, and a growing overlap between the two. I run the legal function at FaceApp, a consumer product used by hundreds of millions, across IP, privacy, corporate, and app-store disputes. Before that, KPMG and senior in-house roles at international tech companies across Cyprus, the EU, and beyond.
Legal engineering is what I genuinely love. I've built RAG pipelines over our internal corpus, contract analytics, MCP servers that keep client data off the API. I work with prompt engineering and fine-tuning, and I deploy local models when client data shouldn't reach a hosted one in the first place.
I've also shipped legal-tech projects as a product owner, taking ideas from zero to launch and owning the spec, the flows, and the API.
I teach a course and run a Telegram channel for lawyers learning to work with LLMs — 100+ students so far.
Tools and workflows I build for legal teams — in-house and law firms.
PII Shield
Anonymize documents before Claude sees them. Restore real data after analysis.
MCP server for Claude Desktop. Reads documents locally, swaps every name, ID, and number for placeholders (<PERSON_1>, <ORG_1>, etc.), and sends only the anonymized text to Claude.
After analysis, the real values are stitched back into the final document — all on your machine. PII never enters the API.
- Node.js
- MCP
- GLiNER
- ONNX
- DOCX
memoforge
From one legal question to a finished .docx memo.
Cowork plugin that mirrors how a small legal team produces a memo. A triage analyst clarifies missing facts. Three researchers pull statutes, case law, and regulator guidance in parallel. A writer drafts IRAC. Five reviewers stress-test it. A mediator consolidates findings. The writer revises — up to three rounds — before polish and .docx export.
Ships with the full source pack, per-reviewer findings, and an honest verdict — approved, forced_exit_on_v3, or manual_review_required. No false confidence.
Optional Style Studio turns your firm's example memos into a saved house-style profile.
- Cowork plugin
- 15 agents
- IRAC
- MCP citations
- python-docx
Brand monitor
Find the apps stealing your name and the sites cracking your build — every week, on autopilot.
Skill that scans the App Store and the open web for knockoffs and cracked APKs of your app. iTunes Search across countries plus fuzzy-matching narrows ~1,000 listings to ~50 real candidates. Icon vision and WHOIS-backed evidence packages confirm the rest. An LLM layer scores risk, recommends action, and ships a prioritized Excel report.
The whole thing is a single SKILL.md — drop it into Claude or OpenClaw, tell it your trademark and target countries, and it runs itself. Hours of manual work → ~20 minutes of review.
- SKILL.md
- iTunes API
- Vision
- WHOIS
- Excel
Notes on legal-AI, originally posted on LinkedIn.
Subagents vs Agent Teams in Claude
Two ways to spawn agents in your Claude subscription — Subagents (orchestrated, parallel, scored to 90+) and Agent Teams (independent, messaging each other, watched live).
Free knowledge base your team's Claude can read
A free shared base for the team — Google Drive + Obsidian + Claude Desktop's Drive MCP. Per-folder indexes keep retrieval fast; add a colleague to Drive and they have the same base.
PII Shield v2 — anonymize the matter, not the document
v2 keeps placeholders consistent across an entire matter — 10 contracts as one session, encrypted handoff with a passphrase, in-chat entity review.