Janady13/AI_Brain / daemon/datasets/knowledge_workflows.json

1[ 2 { 3 "name": "Campaign Operations RAG Pipeline", 4 "category": "rag", 5 "description": "Build a retrieval-augmented knowledge base that keeps campaign playbooks, case studies, and benchmarks instantly queryable for the marketing ANI.", 6 "trigger_keywords": [ 7 "rag", 8 "retrieval", 9 "knowledge", 10 "playbook", 11 "case", 12 "faiss", 13 "vector", 14 "context", 15 "document" 16 ], 17 "indexing_steps": [ 18 "Ingest marketing collateral (case studies, campaign briefs, sales enablement) via document loaders.", 19 "Split documents with a recursive character splitter tuned to 1.5k token windows to respect LLM context limits.", 20 "Generate embeddings with a multilingual MiniLM family model and store vectors plus chunk metadata.", 21 "Backfill metadata including campaign channel, audience, funnel stage, and performance tags for filtering.", 22 "Schedule nightly refresh jobs that append new collateral and recompute impacted embeddings." 23 ], 24 "retrieval_steps": [ 25 "Embed the incoming marketing brief using the same encoder as the index stage.", 26 "Run a similarity search capped at 8 chunks with score thresholding to prevent noisy context.", 27 "Augment the LLM prompt with deduplicated context snippets ordered by funnel stage relevance.", 28 "Cache the enriched prompt and response pair inside a lightweight SQLite cache for reuse across similar briefs.", 29 "Log citations for every retrieved chunk so operators can audit the supporting evidence." 30 ], 31 "vector_store": "faiss", 32 "embedding_models": [ 33 "sentence-transformers/all-MiniLM-L6-v2", 34 "openai/text-embedding-3-small" 35 ], 36 "tooling": [ 37 "LangChain", 38 "LlamaIndex", 39 "Haystack", 40 "Chroma" 41 ], 42 "evaluation": [ 43 "context_precision", 44 "answer_recall", 45 "citation_coverage" 46 ], 47 "governance": [ 48 "versioned document manifests", 49 "PII scrubbing on ingest", 50 "automated relevance regression tests" 51 ], 52 "caching": [ 53 "persona prompt cache", 54 "embedding warm-start cache" 55 ], 56 "source": "knowledge_workflows_dataset" 57 }, 58 { 59 "name": "Revenue Intelligence RAG Accelerator", 60 "category": "rag", 61 "description": "Operationalize a RAG loop dedicated to monetization patterns so the ANI can answer pricing, checkout, and ROI prompts instantly.", 62 "trigger_keywords": [ 63 "stripe", 64 "paypal", 65 "checkout", 66 "pricing", 67 "monetization", 68 "revenue", 69 "catalog" 70 ], 71 "indexing_steps": [ 72 "Harvest pricing experiments, checkout flows, and monetization FAQs from Stripe and PayPal knowledge bases.", 73 "Chunk data by workflow (trial, subscription, invoice) and annotate success metrics like conversion lift and churn reduction.", 74 "Embed chunks with financial-tuned embeddings and store in a vector database with metadata filters for payment provider.", 75 "Link atomic facts to the statistical reasoning store so live metrics can adjust retrieval priority.", 76 "Automate delta detection on provider changelog feeds and enqueue re-embeddings when new payment primitives launch." 77 ], 78 "retrieval_steps": [ 79 "Detect the monetization provider mentioned in the brief and prioritize matching metadata in similarity search.", 80 "Blend top-k retrieval with hybrid keyword filtering to surface compliance or payout specific notes.", 81 "Assemble augmented prompts with explicit instructions to cite conversion benchmarks and payout cadences.", 82 "Persist high performing augmented prompts into a cache-augmented generation layer for rapid reuse.", 83 "Route unresolved gaps into a follow-up task list for live analyst review to maintain coverage." 84 ], 85 "vector_store": "weaviate", 86 "embedding_models": [ 87 "cohere/embed-english-v3.0" 88 ], 89 "tooling": [ 90 "Pinecone", 91 "Airbyte", 92 "dbt", 93 "LangChain" 94 ], 95 "evaluation": [ 96 "prompt_groundedness", 97 "revenue_accuracy", 98 "latency_ms" 99 ], 100 "governance": [ 101 "payment compliance review", 102 "anomaly detection dashboards" 103 ], 104 "caching": [ 105 "checkout flow blueprint cache" 106 ], 107 "source": "knowledge_workflows_dataset" 108 }, 109 { 110 "name": "Marketing Knowledge Graph", 111 "category": "kag", 112 "description": "Construct a domain knowledge graph that links campaigns, audiences, offers, and outcomes so the ANI can reason over multi-hop relationships.", 113 "trigger_keywords": [ 114 "kag", 115 "graph", 116 "knowledge", 117 "entity", 118 "relationship", 119 "ontology" 120 ], 121 "indexing_steps": [ 122 "Extract entities (campaign, audience, offer, channel, KPI) from curated collateral using spaCy patterns.", 123 "Map relationships like 'targets', 'drives', and 'depends_on' into a property graph schema.", 124 "Enrich nodes with statistical priors (ROI, CTR, confidence) from the benchmarks dataset.", 125 "Back-link every node to source documents for traceability and citation.", 126 "Publish the graph into Neo4j and expose a Cypher API alongside GraphQL wrappers." 127 ], 128 "retrieval_steps": [ 129 "Translate natural language questions into Cypher using a prompt template tuned for marketing ontology.", 130 "Run graph traversals to collect multi-hop facts and supporting documents.", 131 "Fuse graph answers with semantic snippets retrieved via hybrid vector search for narrative color.", 132 "Score candidate answers for factual consistency before handing context to the generator.", 133 "Record provenance (node ids, relationship ids, source docs) for downstream auditing." 134 ], 135 "vector_store": "neo4j-graph-data-science", 136 "embedding_models": [ 137 "openai/text-embedding-3-large", 138 "graphsage" 139 ], 140 "tooling": [ 141 "Neo4j", 142 "AWS Neptune", 143 "LlamaIndex", 144 "ArangoDB" 145 ], 146 "evaluation": [ 147 "multi-hop_accuracy", 148 "graph_coverage", 149 "consistency_score" 150 ], 151 "governance": [ 152 "schema governance board", 153 "entity resolution thresholds", 154 "PII lineage tracking" 155 ], 156 "caching": [ 157 "subgraph cache for frequent buyer journeys" 158 ], 159 "source": "knowledge_workflows_dataset" 160 }, 161 { 162 "name": "Financial Knowledge Graph", 163 "category": "kag", 164 "description": "Blend stock, crypto, and marketplace data into a finance-grade knowledge graph that the ANI can query for revenue forecasting and risk controls.", 165 "trigger_keywords": [ 166 "stock", 167 "crypto", 168 "market", 169 "risk", 170 "portfolio", 171 "trading", 172 "forecast" 173 ], 174 "indexing_steps": [ 175 "Stream CoinGecko, CoinCap, and global equity data into staging tables with hourly cadence.", 176 "Normalize tickers, trading pairs, and asset metadata, linking to benchmarks and scenario priors.", 177 "Create graph nodes for assets, indicators, risk regimes, and revenue programs with timestamped properties.", 178 "Derive relationships capturing lead/lag correlations, volatility clusters, and revenue sensitivities.", 179 "Expose the knowledge graph through Amazon Neptune Streams so downstream agents can subscribe to changes." 180 ], 181 "retrieval_steps": [ 182 "Generate Cypher queries that traverse from marketing prompts to correlated assets and risk regimes.", 183 "Combine graph answers with statistical reasoning summaries for sanity checks.", 184 "Return structured responses containing projected revenue multipliers and hedging recommendations.", 185 "Persist high-value traversals into a cache-augmented graph lookup table for millisecond reuse.", 186 "Alert operators when new correlations appear so scenario planning rules can be updated." 187 ], 188 "vector_store": "aws-neptune", 189 "embedding_models": [ 190 "fastText", 191 "node2vec" 192 ], 193 "tooling": [ 194 "Amazon Neptune", 195 "Neo4j", 196 "Apache Kafka", 197 "LangChain" 198 ], 199 "evaluation": [ 200 "forecast_accuracy", 201 "hedge_recall", 202 "latency_ms" 203 ], 204 "governance": [ 205 "financial compliance review", 206 "anomaly monitoring", 207 "disaster recovery runbooks" 208 ], 209 "caching": [ 210 "market regime cache" 211 ], 212 "source": "knowledge_workflows_dataset" 213 } 214] 215