Pipeless Recommendations

@pipeless show me personalized article recommendations for user_12345 based on their reading history and follows

Based on user_12345's activity, here are 8 recommended articles: 1. "Scaling Microservices with Kubernetes" (score: 0.87) — matches reading patterns in infrastructure topics 2. "React Server Components Deep Dive" (score: 0.82) — author followed by user 3. "Database Indexing Strategies" (score: 0.79) — collaborative signal from similar readers 4. "TypeScript 5.3 Features" (score: 0.76) 5. "Observability Best Practices" (score: 0.74) Recommendations exclude 2 articles user_12345 previously dismissed. Scores reflect combined signals from positive interactions (reads, likes), follow relationships, and tag preferences.

This example demonstrates Pipeless's core recommendation engine, which synthesizes multiple signal types (positive interactions, follows, dismissals) into ranked suggestions. Requires prior event ingestion — recommendations improve as you record more user activity. The MCP returns bounded lists (typically 10-50 items), not infinite feeds.

@pipeless record that user_7890 just read article_456 and spent 3 minutes on it, then create a 'like' event for the same article

Recorded 2 events for user_7890: 1. Event created: user_7890 → 'read' → article_456 (timestamp: 2025-05-15T14:32:10Z, metadata: duration=180s) 2. Event created: user_7890 → 'like' → article_456 (timestamp: 2025-05-15T14:35:22Z) Both events are now feeding Pipeless's recommendation algorithms. Object records for user_7890 and article_456 have been created or updated. Note: deleting these events later will not remove the object records themselves.

This write-heavy example shows how you feed behavioral data into Pipeless. The MCP supports single-event and batch creation (up to 10 events per batch). Critical detail: object records persist even after event deletion, so plan your data retention strategy accordingly. Events must include start object, relationship type, and end object.

@pipeless find content similar to article_789 using collaborative filtering, then summarize why each recommendation matches

Found 6 related articles to article_789 ("Building Resilient APIs"): 1. "Circuit Breaker Patterns" — 47 users who read article_789 also engaged with this 2. "Rate Limiting Strategies" — shares 'api-design' and 'reliability' tags 3. "Retry Logic Best Practices" — collaborative signal from 32 overlapping readers 4. "Distributed Tracing Guide" — tag overlap: 'observability', 'microservices' Pipeless used both tag-based similarity and collaborative interaction signals (users who read A also read B). These related-content results don't require per-user personalization, making them useful for "more like this" widgets.

This synthesis example pairs Pipeless's related-content tool with AI reasoning to explain recommendation logic. Unlike personalized recommendations, related-content queries are object-to-object (no user context needed). Useful for discovery features. The MCP returns raw scores; the AI interprets signals (tags, collaborative filtering) into human-readable explanations.

When Pipeless wins for personalized doc discovery in small teams

A 6-person ops team maintains 200+ runbooks in Notion and wants each engineer to see relevant docs without manual tagging. Pipeless tracks who views what, then surfaces related content based on collaborative signals—if three people who read the same incident guide also read a specific deployment doc, that doc bubbles up for the fourth. The List Recommended Content tool handles this with follow and tag signals, and the activity feed keeps a log of what's trending. This works until you hit 50+ active users or need real-time updates; Pipeless batches events and recommendations run async, so there's a 5-10 minute lag. If your team needs instant feed refresh or you're over 100 docs with complex taxonomy, you'll outgrow this fast. For a small team with stable content and weekly discovery needs, Pipeless delivers personalized recommendations without building your own rec engine.

Using Pipeless to surface help docs during live support chats

A 3-person support team handles 40 tickets a day and wants agents to see relevant help articles as they type responses. Pipeless records each article view as an event (Create Event tool), then uses List Related Content to suggest docs based on what other agents referenced for similar issues. The collaborative-interaction signal means if two agents both viewed Article A and Article B while resolving password-reset tickets, Article B will surface when the third agent opens a similar ticket. This breaks down if your article corpus is under 50 pieces—there's not enough signal for meaningful recommendations—or if you need sub-second latency, since Pipeless processes events in batch windows. For a small team with 100+ articles and agents who can wait 10 seconds for suggestions, this replaces manual search and cuts resolution time. Larger teams should look at dedicated support-AI tools with tighter integrations.

When Pipeless fits for deal-stage playbook surfacing

A 5-rep sales team has 30 playbooks (pitch decks, objection-handling scripts, case studies) and wants each rep to see the right assets as deals progress. Pipeless tracks which playbooks a rep views at each stage (Create Events Batch for bulk logging), then recommends content using List Recommended Content with author and tag filters—if a rep is in discovery and three peers viewed the same competitor comparison doc, it surfaces automatically. The activity feed (List Activity Feed) shows what top performers are reading, creating a passive coaching loop. This falls apart if your playbook library is under 20 items or if reps need recommendations inside Salesforce without switching contexts; Pipeless has no native CRM integration, so reps must open a separate interface. For a small team with a standalone playbook repo and reps who check it daily, Pipeless builds a lightweight recommendation layer without custom ML work.