Vlastní data (RAG) pro vašeho AI agenta část třetí - AI samo volí strategii, jak se dostat k datům

Question: I have seen all Star Wars movies and would like tips for something similar I can watch next.
Generating sub-questions.
  -> Sub-question: What are some science fiction movies similar to Star Wars?
  -> Sub-question: Can you suggest movies or shows with epic space battles and adventures?
  -> Sub-question: What are some recommendations for movies with themes of rebellion against tyranny?
  -> Sub-question: Are there any film series with iconic characters and expansive universes like Star Wars?
  -> Sub-question: What are other popular franchises that mix science fiction and fantasy elements?

Your task is to rephrase the user's question to improve retrieval accuracy in a Retrieval Augmented Generation system, following a ReSP pattern (retrieve, summarize, plan).

Keep the following in mind:
- Preserve all essential details from the original question.
- Do not introduce new information or context.
- Improve phrasing to be concise and clear.

Your task is to create a query strategy for a Retrieval Augmented Generation system. The strategy should:

# Instructions
- Analyze the user query and decide whether to use semantic search, full_text search or graph search.
- Identify one or more traits (e.g., Movie, Genre, Character, Setting, Theme, Series) to target.
- Rewrite the user query:
    - For keyword search, extract ONLY the essential identifying terms (names, entities, specific descriptors) from the user's question.
    - For semantic search, rewrite the user query to semantically match articles that might be in the knowledge base.
- Provide one, preferably two or more strategies. Make sure each strategy is distinct.
- All SQL searches must return no more than 20 results.
- For graph search, use the MATCH clause to find related nodes and edges.
- For keyword search, use CONTAINS clause to find related nodes and edges.
- Use TOP 20 in the final result to limit the number of results returned.

-- Similarity search on Movie
SELECT TOP 10 Id, Content, vector_distance('cosine', Embedding, @q) AS Similarity
FROM [dbo].[Movie]
WHERE Embedding IS NOT NULL AND Content IS NOT NULL
ORDER BY Similarity ASC

-- Fulltext search on Movie together with semantic search on results
SELECT TOP 10 Id, Content, vector_distance('cosine', Embedding, @q) AS Similarity
FROM [dbo].[Movie]
WHERE CONTAINS (Content, '"Indiana Jones" AND "Temple"')
ORDER BY Similarity ASC

-- Semantic search on Character, graph search to find related Movie
WITH topcharacters AS (
    SELECT TOP 50 Id, vector_distance('cosine', Embedding, @q) AS Similarity
    FROM [dbo].[Character]
    WHERE Embedding IS NOT NULL AND Content IS NOT NULL
    ORDER BY Similarity ASC
),
MovieMatches AS (
    SELECT TOP 10 m.Content as Content, COUNT(c.Id) AS MatchedCharacterCount
    FROM [dbo].[Movie] m, [dbo].[FEATURES_CHARACTER] fc, [dbo].[Character] c
    WHERE MATCH (m-(fc)->c)
    AND c.Id IN (SELECT Id FROM topcharacters)
    GROUP BY m.Content
)
SELECT TOP 10 Content
FROM MovieMatches
ORDER BY MatchedCharacterCount DESC;

class QueryStrategy(BaseModel):
    id: str = Field(
        description="Unique random identifier for this strategy"
    )
    message_to_user: str = Field(
        description="Message to be displayed to the user, explaining the purpose of this query. Include type of each query step and its relation to other steps. Wording should be about what you are doing, example: 'Searching Movie using keyword search for Star Wars followed by graph query to get related genres and semantic movie search among results."
    )
    query: str = Field(
        description="Rewrite query to better match knowledge base using semantic search. This field is used to create embedding and store as @q variable for command to use."
    )
    command: str = Field(
        description="Command to be executed in the database. This should be a SQL query that can be executed directly against the database. All text-based queries should be hardcoded, just vector will be referenced as %q."
    )

class QueryStrategies(BaseModel):
    strategies: List[QueryStrategy] = Field(
        description="List of strategies, each containing a series of steps to be executed sequentially. Each strategy is a sequence of steps, where each step depends on the results of previous steps."
    )

Your task is to fix a Microsoft SQL query that is failing with syntax errors.

# Instructions
- Analyze the query and errors
- All SQL searches must return no more than 20 results. Use various techniques to select best ones if query matches more, eg. reorder results using semantic search (vector similarity).
- For graph search, use the MATCH clause to find related nodes and edges. Use WITH clause to create temporary tables for intermediate results.
- Use TOP 20 in the final result to limit the number of results returned.

# Original query
{{query}}

# Error
{{error}}

Generating query strategies.
 -> Using semantic search to find science fiction movies similar to Star Wars.
    -> Found 20 records
 -> Performing keyword search for science fiction movies and Star Wars, followed by semantic search within results.
    -> Trying to fix error in query
    -> Fixed query successful on attempt 1.
    -> Found 20 records
 -> Searching for Movies and Series using semantic search to match the theme of epic space battles and adventures.
    -> Found 20 records
 -> Searching for Movies and Series using semantic search to match the theme of epic space battles and adventures.
    -> Found 20 records
 -> Searching for themes related to rebellion against tyranny using semantic search, followed by a graph query to retrieve associated movies.
    -> Trying to fix error in query
    -> Fixed query successful on attempt 4.
    -> Found 20 records
 -> Performing a semantic search directly on movies to find content matching the theme of rebellion against tyranny.
    -> Found 20 records
 -> Performing semantic search for expansive film series with iconic characters and universes.
    -> Found 20 records
 -> Using keyword search for iconic series and characters, followed by graph search to explore related movies.
    -> Found 18 records
 -> Performing semantic search on the Series table to identify franchises blending science fiction and fantasy elements.
    -> Found 20 records
 -> Using a keyword-based search on the Theme table to find themes combining science fiction and fantasy, followed by graph search to identify related franchises.
    -> Trying to fix error in query
    -> Fixed query successful on attempt 1.
    -> Found 20 records

# Role
Your task is to review JSON with executions of different query strategies each identified by unique ID, assess results of each strategy and select the one that is most useful for answering the user question. Based on the selected strategy, you will tak winning SQL query and modify it capture larger set of results by increasing TOP value to 60.

# Instructions
- Review the JSON with query strategy executions and their results. JSON will be provided to you in the next user message.
- User question will be provided to you in the last user message as a string.
- Assess quality of results for each strategy. Consider the following:
  - Are the results relevant to the user question?
  - Are there any missing or incomplete results?
  - Is the number of results appropriate?
- Based on your assessment, select the best strategy.
- Modify the winning SQL query to increase the TOP value to 60.
- Provide the modified SQL query as output.

Selecting the best strategy based on the results.
Expanding results of the best strategy.
 -> This query is designed to find film series that are similar to Star Wars in their blending of science fiction, fantasy, and epic storytelling. It aims to identify options for viewers seeking engaging narratives and expansive universes.
    -> Found 60 records

If you've watched all the "Star Wars" movies and are seeking something similar, here are recommendations based on themes, world-building, and narrative structures akin to the "Star Wars" saga:

### 1. "Star Trek" Movie Series

- Why Similar: Like "Star Wars," the "Star Trek" series explores interstellar conflicts, diplomacy, and the moral complexities of leadership and humanity's place in the universe. It features compelling ensemble casts, advanced technology, and philosophical undertones.
- Highlights: Iconic characters like Captain Kirk and Spock navigate moral dilemmas and thrilling adventures across the galaxy. Key films include Star Trek II: The Wrath of Khan and the 2009 reboot Star Trek, which reintroduce classic elements with modern storytelling.

### 2. "Battlestar Galactica" Series

- Why Similar: This series delves into the survival of humanity against an AI-driven race, the Cylons, blending space-based action with moral struggles and complex interpersonal dynamics. It mirrors "Star Wars" themes of rebellion and redemption.
- Highlights: The movies Battlestar Galactica: Razor and Battlestar Galactica: The Plan provide deeper insights into character backstories and philosophical questions about existence and war.

### 3. "Guardians of the Galaxy" Series

- Why Similar: With its mix of humor, heartfelt character dynamics, and intergalactic adventures, "Guardians of the Galaxy" mirrors the camaraderie and action-packed escapades of "Star Wars."
- Highlights: The films Guardians of the Galaxy and Guardians of the Galaxy Vol. 2 focus on a ragtag group of misfits who band together to confront cosmic threats, offering both emotional depth and humor.

### 4. "Pacific Rim" Series

- Why Similar: Lush visuals, teamwork, and humanity's battle against towering alien threats resonate with the larger-than-life battles in "Star Wars."
- Highlights: Pilots join forces to control giant robots in Pacific Rim and Pacific Rim: Uprising, blending epic battles with themes of sacrifice and resilience.

### 5. "Dune Saga"

- Why Similar: "Dune" shares the epic scale and political intrigue of "Star Wars," with a focus on power and rebellion in a world driven by mystical elements similar to the Force.