raheesahmed/sajicode/ai-engineer

Documents → Chunking → Embedding → Vector Store
                ↓
         Metadata extraction
         (title, source, date)

Query → Query Understanding → Retrieval → Reranking → Generation
              ↓                    ↓           ↓
         Expansion/       Hybrid search    Cross-encoder
         Decomposition    (vector + BM25)   scoring

// Recursive text splitter — best default
const splitter = new RecursiveCharacterTextSplitter({
  chunkSize: 1000,
  chunkOverlap: 200,
  separators: ["\n\n", "\n", ". ", " ", ""],
});
// Semantic chunking — for high-quality retrieval
const semanticSplitter = new SemanticChunker(embeddings, {
  breakpointThresholdType: "percentile",
  breakpointThresholdAmount: 95,
});

import { StateGraph, MessagesAnnotation } from "@langchain/langgraph";
import { ToolNode } from "@langchain/langgraph/prebuilt";
const agentNode = async (state: typeof MessagesAnnotation.State) => {
  const response = await model.invoke(state.messages);
  return { messages: [response] };
};
const shouldContinue = (state: typeof MessagesAnnotation.State) => {
  const lastMessage = state.messages[state.messages.length - 1];
  return lastMessage.tool_calls?.length ? "tools" : "__end__";
};
const graph = new StateGraph(MessagesAnnotation)
  .addNode("agent", agentNode)
  .addNode("tools", new ToolNode(tools))
  .addEdge("__start__", "agent")
  .addConditionalEdges("agent", shouldContinue)
  .addEdge("tools", "agent")
  .compile();

const supervisorNode = async (state: AgentState) => {
  const response = await supervisorModel.invoke([
    { role: "system", content: "Route to the right specialist agent." },
    ...state.messages,
  ]);
  return { next: response.content }; // "researcher" | "coder" | "reviewer"
};

const schema = z.object({
  sentiment: z.enum(["positive", "negative", "neutral"]),
  confidence: z.number().min(0).max(1),
  reasoning: z.string(),
});
const structuredLlm = model.withStructuredOutput(schema);
const result = await structuredLlm.invoke("Analyze: Great product!");

You are an expert analyst. Think through this step by step:
 
1. First, identify the key entities in the text
2. Then, determine the relationships between them
3. Finally, synthesize your findings into a structured answer
 
Text: {input}

const fewShotPrompt = ChatPromptTemplate.fromMessages([
  ["system", "Extract structured data from text."],
  ["human", "John works at Google since 2020"],
  ["ai", '{"name": "John", "company": "Google", "year": 2020}'],
  ["human", "Sarah joined Meta in 2023"],
  ["ai", '{"name": "Sarah", "company": "Meta", "year": 2023}'],
  ["human", "{input}"],
]);

const cache = new SemanticCache({
  embeddings,
  vectorStore,
  similarityThreshold: 0.95,
});
async function cachedInvoke(prompt: string) {
  const cached = await cache.lookup(prompt);
  if (cached) return cached;
  const result = await model.invoke(prompt);
  await cache.store(prompt, result);
  return result;
}