Home
Scenario-based tutorials with real screenshots, prompts, workflow templates and pitfalls.
Breaks RAG quality into three measurable layers-retrieval quality (context precision/recall), generation faithfulness, and answer relevance-and evaluates each with Ragas, TruLens, and DeepEval. Includes eval-set field design, 4 copy-paste code/prompt blocks, 6 pitfalls, and 5 FAQs. Complements the existing "Dify build SOP" by focusing purely on evaluation.
Breaks AI music production into five steps-concept, lyrics, AI generation (Suno/Udio), post-production, and distribution-with real tools, three copy-paste Prompts, six pitfalls, and five FAQs for a complete zero-to-release pipeline.
A one-person, zero-recording podcast pipeline across topic, script, TTS synthesis, polish, editing, and distribution, with 3 copy-paste prompts, pitfalls, and FAQ.
From installing ComfyUI, placing the right checkpoint, building a base txt2img workflow with a fixed seed for reproducibility, to layering in LoRA for style and IPAdapter for reference-image transfer, a five-step SOP to achieve consistent style and character across generations, with real commands, a KSampler parameter table, workflow JSON, pitfalls, and FAQ.
Funnel multi-channel customer feedback through one n8n pipe: auto-dedupe + PII mask, LLM tags type/sentiment/urgency/module, bugs go straight to GitHub Issues, P0 pushes WeCom with @mention, rest land in a summary table for weekly trend analysis. Includes .json template + classification prompt + 7 pitfalls (PII masking red line).
Feed one source article into n8n; the LLM rewrites it per X/Xiaohongshu/WeChat/Zhihu specs and archives all four to Feishu/Notion for human sign-off. Includes n8n JSON template, four platform-specific prompts, and six pitfalls.
A six-step prompt pack turning vague requirements into build-ready specs: interview to surface constraints, user stories to draw boundaries, functional structure for the skeleton, technical constraints to prevent blowups, testable acceptance cases, and review iteration to find holes. Works with Claude/DeepSeek, pairs with the /spec skill.
A six-step prompt pipeline that turns podcast transcripts into structured knowledge cards-preprocessing, insight extraction, quote cards, action checklists, knowledge graph nodes, and critical review. Pair with Whisper/Tingwu for transcription, feed to Claude/DeepSeek long-context models.
A 6-step data cleaning method where the LLM provides judgment and Python provides execution: profile dirty points, define rules, script execution, LLM for hard rules, validate, and audit trail. Includes 3 copy-paste prompts and 5 pitfalls.
Refactoring legacy code with Claude Code/Cursor takes more than a "refactor this" prompt. This 5-step SOP hard-codes engineering discipline: build a test safety net first, define refactoring boundaries, give the AI precise context, execute step-by-step with manual diff review, then regression-verify. Includes 4 copy-ready prompt templates and 5 real-world pitfalls.