Most people remember workplace events incompletely or with details shifted by time and emotion, while AI systems that can retrieve and cross-reference your actual documented records create a reliable external memory—what actually happened according to what you wrote at the time, not what you remember later. This distinction matters enormously if you ever need to defend yourself against mischaracterizations.
Retrieval-Augmented Generation (RAG) is a technical pattern where an AI system searches through a knowledge base to find relevant information, then uses that information to answer your question more accurately. For workplace documentation, RAG transforms scattered notes and emails into an intelligent searchable archive that surfaces patterns you'd otherwise miss.
You have hundreds of emails, Slack messages, performance reviews, and meeting notes. When your manager gaslights you about something that happened six months ago, you can't instantly recall the exact documentation. Even if you search your email, you might find the evidence but miss the broader pattern—this is the seventh time they've made a similar claim.
RAG systems solve this by indexing all your documentation and allowing semantic search. Instead of searching for exact keywords, you ask: "When has my manager disputed my work quality?" The system retrieves all relevant instances, and an AI synthesizes them into a response showing the pattern.
First, you upload or connect your documentation sources: Gmail, Slack, meeting transcripts from Otter.ai, notes in Notion. The RAG system breaks this into chunks and creates embeddings—mathematical representations of meaning. This happens once and stays indexed.
When you ask a question, the RAG system searches the embeddings for relevant chunks (retrieval), then feeds those chunks to an LLM like Claude with your question (generation). The LLM synthesizes the retrieved documents into an answer.
Example: You ask, "What evidence exists that I raised concerns about the project scope before the deadline passed?" The RAG system retrieves every message where you mentioned scope concerns, presents them to Claude in chronological order, and Claude generates a summary showing the timeline and evidence.
The technical implementation varies, but the concept is consistent. You can use Claude's built-in file search functionality (upload PDF archives of important emails), or build more sophisticated systems using Notion as your indexed knowledge base.
Best practice: Maintain a Notion database where you regularly add new workplace events with consistent structure—date, participants, summary, source document, relevance tags. This becomes your RAG corpus. When you need to query it, Claude can search the Notion archive and synthesize patterns.
RAG can surface relevant information but doesn't automatically maintain temporal context. If you ask "When did my manager approve my promotion?" the RAG might return both the actual approval email and a later message where they questioned whether they'd approved it. The AI must rank these chronologically and note the contradiction—something RAG systems sometimes miss without explicit instruction.
Mitigate this by tagging documents with temporal markers ("approval", "contradiction", "denial") so RAG searches return context about how claims have evolved over time.
RAG systems that use your company's infrastructure (or cloud services) create data residency questions. If you're using a self-hosted RAG implementation, your data stays under your control. If you're relying on third-party cloud RAG services, review their data retention policies.
For legally sensitive documentation (particularly in retaliation or discrimination cases), you may want local RAG implementations rather than cloud-dependent solutions. This is more technically complex but gives you full control over where your evidence lives.
Keyword search for "performance issues" might miss a manager's message that says "your execution has been inconsistent." RAG's semantic search understands these are conceptually similar and retrieves both. This is powerful for documenting patterns because gaslighting often involves the same underlying pattern expressed with different vocabulary.
Ask RAG: "Find all instances where my manager questioned my technical capability." It returns messages about "inconsistent execution", "missed details", "didn't understand the scope"—semantically similar claims that together show a pattern of undermining.
Start simple: Create a Notion database of important workplace events. Tag them consistently (topic, date, participants, sentiment). Use Claude to query it with questions. As you scale, consider more sophisticated RAG tools, but basic Notion + Claude semantic search covers most workplace documentation needs.
Try this: Export 3 months of emails from a difficult working relationship. Create a Notion database with columns for Date, From, Key Claims, Your Response, and Tags. Add 15-20 significant messages. Then ask Claude: "Using this Notion database, show me every instance where my manager questioned my work quality." The AI will surface patterns across your documentation that you probably didn't consciously notice. This is RAG in its simplest form—and it's remarkably powerful for identifying gaslighting patterns.
Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.
Explore related journeys or tell Peri what you're working through.