rag

Name: rag
Author: giuseppe-trisciuoglio

1.6kinstalls36forks

How to install rag?

giuseppe-trisciuoglio/developer-kit/rag

$npx -y skills add giuseppe-trisciuoglio/developer-kit --skill rag

Installs into the current project.

Summary

Implements document chunking, embedding generation, vector storage, and retrieval pipelines for Retrieval-Augmented Generation systems. Use when building RAG applications, creating document Q&A systems, or integrating AI with knowledge bases.

Files · 1

View on GitHub

SKILL.md

1	# RAG Implementation
2
3	Build Retrieval-Augmented Generation systems that extend AI capabilities with external knowledge sources.
4
5	## Overview
6
7	This skill covers: document processing, embedding generation, vector storage, retrieval configuration, and RAG pipeline implementation.
8
9	## When to Use
10
11	- Building Q&A systems over proprietary documents
12	- Creating chatbots with factual information from knowledge bases
13	- Implementing semantic search with natural language queries
14	- Reducing hallucinations with grounded, sourced responses
15	- Building documentation assistants and research tools
16	- Enabling AI systems to access domain-specific knowledge
17
18	## Instructions
19
20	### Step 1: Choose Vector Database
21
22	Select based on your requirements:
23
24	\| Requirement \| Recommended \|
25	\|-------------\|-------------\|
26	\| Production scalability \| Pinecone, Milvus \|
27	\| Open-source \| Weaviate, Qdrant \|
28	\| Local development \| Chroma, FAISS \|
29	\| Hybrid search \| Weaviate with BM25 \|
30
31	### Step 2: Select Embedding Model
32
33	\| Use Case \| Model \|
34	\|----------\|-------\|
35	\| General purpose \| text-embedding-ada-002 \|
36	\| Fast and lightweight \| all-MiniLM-L6-v2 \|
37	\| Multilingual \| e5-large-v2 \|
38	\| Best performance \| bge-large-en-v1.5 \|
39
40	### Step 3: Implement Document Processing Pipeline
41
42	1. Load documents from source (file system, database, API)
43	2. Clean and preprocess (remove formatting, normalize text)
44	3. Split documents into chunks with appropriate strategy
45	4. Generate embeddings for each chunk
46	5. Store embeddings in vector database with metadata
47
48	Validation: Verify embeddings were generated successfully:
49	```java
50	List<Embedding> embeddings = embeddingModel.embedAll(segments);
51	if (embeddings.isEmpty() \|\| embeddings.get(0).dimension() != expectedDim) {
52	throw new IllegalStateException("Embedding generation failed");
53	}
54	```
55
56	### Step 4: Configure Retrieval Strategy
57
58	Choose the appropriate strategy:
59
60	- Dense Retrieval: Semantic similarity via embeddings (default for most cases)
61	- Hybrid Search: Dense + sparse retrieval for better coverage
62	- Metadata Filtering: Filter by document attributes
63	- Reranking: Cross-encoder reranking for high-precision requirements
64
65	### Step 5: Build RAG Pipeline
66
67	1. Create content retriever with your embedding store
68	2. Configure AI service with retriever and chat memory
69	3. Implement prompt template with context injection
70	4. Add response validation and grounding checks
71
72	Validation: Test with known queries to verify context injection works correctly.
73
74	Error Handling: For batch ingestion, wrap in retry logic:
75	```java
76	for (Document doc : documents) {
77	int attempts = 0;
78	while (attempts < 3) {
79	try {
80	store.add(embeddingModel.embed(doc).content(), doc.toTextSegment());
81	break;
82	} catch (EmbeddingException e) {
83	attempts++;
84	if (attempts == 3) throw new RuntimeException("Failed after 3 retries", e);
85	}
86	}
87	}
88	```
89
90	### Step 6: Evaluate and Optimize
91
92	1. Measure retrieval metrics: precision@k, recall@k, MRR
93	2. Evaluate answer quality: faithfulness, relevance
94	3. Monitor performance and user feedback
95	4. Iterate on chunking, retrieval, and prompt parameters
96
97	## Examples
98
99	### Example 1: Basic Document Q&A
100
101	```java
102	List<Document> documents = FileSystemDocumentLoader.loadDocuments("/docs");
103
104	InMemoryEmbeddingStore<TextSegment> store = new InMemoryEmbeddingStore<>();
105	EmbeddingStoreIngestor.ingest(documents, store);
106
107	DocumentAssistant assistant = AiServices.builder(DocumentAssistant.class)
108	.chatModel(chatModel)
109	.contentRetriever(EmbeddingStoreContentRetriever.from(store))
110	.build();
111
112	String answer = assistant.answer("What is the company policy on remote work?");
113	```
114
115	### Example 2: Metadata-Filtered Retrieval
116
117	```java
118	EmbeddingStoreContentRetriever retriever = EmbeddingStoreContentRetriever.builder()
119	.embeddingStore(store)
120	.embeddingModel(embeddingModel)
121	.maxResults(5)
122	.minScore(0.7)
123	.filter(metadataKey("category").isEqualTo("technical"))
124	.build();
125	```
126
127	### Example 3: Multi-Source RAG Pipeline
128
129	```java
130	ContentRetriever webRetriever = EmbeddingStoreContentRetriever.from(webStore);
131	ContentRetriever docRetriever = EmbeddingStoreContentRetriever.from(docStore);
132
133	List<Content> results = new ArrayList<>();
134	results.addAll(webRetriever.retrieve(query));
135	results.addAll(docRetriever.retrieve(query));
136
137	List<Content> topResults = reranker.reorder(query, results).subList(0, 5);
138	```
139
140	### Example 4: RAG with Chat Memory
141
142	```java
143	Assistant assistant = AiServices.builder(Assistant.class)
144	.chatModel(chatModel)
145	.chatMemory(MessageWindowChatMemory.withMaxMessages(10))
146	.contentRetriever(retriever)
147	.build();
148
149	assistant.chat("Tell

People who liked this also viewed

6 picks

Type

rag

1.6kinstalls36forks

View on GitHub

How to install rag?

giuseppe-trisciuoglio/developer-kit/rag

$npx -y skills add giuseppe-trisciuoglio/developer-kit --skill rag

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# RAG Implementation
2
3	Build Retrieval-Augmented Generation systems that extend AI capabilities with external knowledge sources.
4
5	## Overview
6
7	This skill covers: document processing, embedding generation, vector storage, retrieval configuration, and RAG pipeline implementation.
8
9	## When to Use
10
11	- Building Q&A systems over proprietary documents
12	- Creating chatbots with factual information from knowledge bases
13	- Implementing semantic search with natural language queries
14	- Reducing hallucinations with grounded, sourced responses
15	- Building documentation assistants and research tools
16	- Enabling AI systems to access domain-specific knowledge
17
18	## Instructions
19
20	### Step 1: Choose Vector Database
21
22	Select based on your requirements:
23
24	\| Requirement \| Recommended \|
25	\|-------------\|-------------\|
26	\| Production scalability \| Pinecone, Milvus \|
27	\| Open-source \| Weaviate, Qdrant \|
28	\| Local development \| Chroma, FAISS \|
29	\| Hybrid search \| Weaviate with BM25 \|
30
31	### Step 2: Select Embedding Model
32
33	\| Use Case \| Model \|
34	\|----------\|-------\|
35	\| General purpose \| text-embedding-ada-002 \|
36	\| Fast and lightweight \| all-MiniLM-L6-v2 \|
37	\| Multilingual \| e5-large-v2 \|
38	\| Best performance \| bge-large-en-v1.5 \|
39
40	### Step 3: Implement Document Processing Pipeline
41
42	1. Load documents from source (file system, database, API)
43	2. Clean and preprocess (remove formatting, normalize text)
44	3. Split documents into chunks with appropriate strategy
45	4. Generate embeddings for each chunk
46	5. Store embeddings in vector database with metadata
47
48	Validation: Verify embeddings were generated successfully:
49	```java
50	List<Embedding> embeddings = embeddingModel.embedAll(segments);
51	if (embeddings.isEmpty() \|\| embeddings.get(0).dimension() != expectedDim) {
52	throw new IllegalStateException("Embedding generation failed");
53	}
54	```
55
56	### Step 4: Configure Retrieval Strategy
57
58	Choose the appropriate strategy:
59
60	- Dense Retrieval: Semantic similarity via embeddings (default for most cases)
61	- Hybrid Search: Dense + sparse retrieval for better coverage
62	- Metadata Filtering: Filter by document attributes
63	- Reranking: Cross-encoder reranking for high-precision requirements
64
65	### Step 5: Build RAG Pipeline
66
67	1. Create content retriever with your embedding store
68	2. Configure AI service with retriever and chat memory
69	3. Implement prompt template with context injection
70	4. Add response validation and grounding checks
71
72	Validation: Test with known queries to verify context injection works correctly.
73
74	Error Handling: For batch ingestion, wrap in retry logic:
75	```java
76	for (Document doc : documents) {
77	int attempts = 0;
78	while (attempts < 3) {
79	try {
80	store.add(embeddingModel.embed(doc).content(), doc.toTextSegment());
81	break;
82	} catch (EmbeddingException e) {
83	attempts++;
84	if (attempts == 3) throw new RuntimeException("Failed after 3 retries", e);
85	}
86	}
87	}
88	```
89
90	### Step 6: Evaluate and Optimize
91
92	1. Measure retrieval metrics: precision@k, recall@k, MRR
93	2. Evaluate answer quality: faithfulness, relevance
94	3. Monitor performance and user feedback
95	4. Iterate on chunking, retrieval, and prompt parameters
96
97	## Examples
98
99	### Example 1: Basic Document Q&A
100
101	```java
102	List<Document> documents = FileSystemDocumentLoader.loadDocuments("/docs");
103
104	InMemoryEmbeddingStore<TextSegment> store = new InMemoryEmbeddingStore<>();
105	EmbeddingStoreIngestor.ingest(documents, store);
106
107	DocumentAssistant assistant = AiServices.builder(DocumentAssistant.class)
108	.chatModel(chatModel)
109	.contentRetriever(EmbeddingStoreContentRetriever.from(store))
110	.build();
111
112	String answer = assistant.answer("What is the company policy on remote work?");
113	```
114
115	### Example 2: Metadata-Filtered Retrieval
116
117	```java
118	EmbeddingStoreContentRetriever retriever = EmbeddingStoreContentRetriever.builder()
119	.embeddingStore(store)
120	.embeddingModel(embeddingModel)
121	.maxResults(5)
122	.minScore(0.7)
123	.filter(metadataKey("category").isEqualTo("technical"))
124	.build();
125	```
126
127	### Example 3: Multi-Source RAG Pipeline
128
129	```java
130	ContentRetriever webRetriever = EmbeddingStoreContentRetriever.from(webStore);
131	ContentRetriever docRetriever = EmbeddingStoreContentRetriever.from(docStore);
132
133	List<Content> results = new ArrayList<>();
134	results.addAll(webRetriever.retrieve(query));
135	results.addAll(docRetriever.retrieve(query));
136
137	List<Content> topResults = reranker.reorder(query, results).subList(0, 5);
138	```
139
140	### Example 4: RAG with Chat Memory
141
142	```java
143	Assistant assistant = AiServices.builder(Assistant.class)
144	.chatModel(chatModel)
145	.chatMemory(MessageWindowChatMemory.withMaxMessages(10))
146	.contentRetriever(retriever)
147	.build();
148
149	assistant.chat("Tell

People who liked this also viewed

6 picks

Type