Custom Prompts with Metadata
Vectara handles the system and user prompts automatically, but if you want to do it yourself, Vectara now empowers developers with a flexible way of customizing prompts with metadata. Our Custom Retrieval Augmented Generation (RAG) Prompt Engine provides several available prompt variables and functions for Scale users to customize prompts.
Available Prompt Variables
The following table shows the available custom prompt variables:
| Variable | Description | Example Usage Input | Example Usage Output |
|---|---|---|---|
| $vectaraOutChars | Number of characters | See below | |
| $vectaraLangCode | ISO639 v3 code for the passed language code | See below | |
| $vectaraQuery | The query provided by the user | Generate a summary in $vectaraOutChars characters in language '${vectaraLangCode}' for the query \"$esc.java(${vectaraQuery})\" solely based on the search results in this chat. | Generate a summary in 512 characters in language 'ara' for the query \"Give me \"some\" search results.\" solely based on the search results in this chat. |
| $vectaraIdxWord | A utility array to convert the index to words i.e "first", "second", "third", "forth", "fifth", "sixth", "seventh", "eighth", "ninth", "tenth" | $vectaraIdxWord[0] | first |
| $vectaraLangName | Set to the requested language name. The language can either be requested explicitly or detected from the language of the query. | You are a helpful assistant. Answer in ${vectaraLangName}. | You are a helpful assistant. Answer in Arabic. |
| $vectaraQueryResults | An array of query results is found in the response, sorted by relevance score. | #foreach ($qResult in $vectaraQueryResults) {"role": "user", "content": "Give me the $vectaraIdxWord[$foreach.index] search result."}, {"role": "assistant", "content": "$esc.java(${qResult.text()})" },#end | {"role": "user", "content": "Give me the second search result."},{"role": "assistant", "content": "2nd result" }, |
Available Prompt Functions
The following table shows the available custom prompt functions:
| Function | Description | Example Usage Input | Example Usage Output |
|---|---|---|---|
| $esc.java(...) | A utility method to escape special charts, has methods such as "esc.java", "esc.url", "esc.xml", "esc.html" | See below | |
| #foreach ($qResult in $vectaraQueryResults) | |||
| $qResult.getText() or $qResult.text() | Returns text of the query result | $qResult.text() | Result text |
| $qResult.docMetadata() | Returns the metadata of the document this result belongs to | $qResult.docMetadata() | {"title": "documentTitle", ...} |
| $qResult.docMetadata().present() | Returns true/false if there are any values present in the metadata | #if ($qResult.docMetadata().present())...#end | |
| $qResult.docMetadata().get("title") | Returns the specified field value from doc metadata, an incorrect key would result in an empty value | $qResult.docMetadata().get("title") | documentTitle |
| $qResult.partMetadata().present() | Returns true/false if there are any values present in the metadata | #if ($qResult.partMetadata().present())...#end | |
| $qResult.partMetadata() | Returns the metadata of the part of the document this result belongs to | $qResult.partMetadata() | {"page": "1", ...} |
| $qResult.partMetadata().get("page") | Returns the specified field value from part metadata, incorrect key would result in empty value | $qResult.docMetadata().get("page") | "1" |
Include Metadata in Prompt
This snippet shows how to get metadata associated with a single result qResult
by retrieving metadata docMetadata from the date that information was
answered answerDate. It then extracts the text content of qResult.
{"role": "assistant", "content": "qResult.docMetadata().get('answerDate') $esc.java(${qResult.getText()})" },
Let's dive into a full custom prompt example that shows more details about a custom prompt with metadata.
Example Custom Prompt for an RFI Answering Bot
The following example prompt creates a Request for information (RFI) answering bot that includes metadata. First, we ask the generative LLM to answer an RFI question and tell it how the results will come back from the query.
We want to iterate through $vectaraQueryResults inserting the results
in the order that we like. $qResult.getText() provides the most relelvant
snippet of text that answers the query from the result. You can iterate to
tell the LLM where to focuse its response, cut or omit results, and tell the
query to reference individual results or even metadata.
For each result, we simulate the user requesting the next search result. Then we have an assistant's response, which includes the answer date from the metadata of the document.
Finally, we generate a comprehensive summary and answer to the question with
additional rules and constraints. For example, if a result does not answer the
question, we do not use the result. If search results are not valid, then the
user gets a response that The returned results did not contain sufficient information to the question.
[
{
"role": "system",
"content": "You are an RFI answering assistant acting on behalf of the company Vectara.
You are provided with search results from previously answered RFIs that may help answer
the given question. The format of each result is the date in which it was answered and
the response text. You must summarize these results into a coherent answer. Only use
information provided in this chat."},
#foreach ($qResult in $vectaraQueryResults)
#if ($foreach.first)
{"role": "user", "content": "Search for \"$esc.java(${vectaraQuery})\", and give me the first search result."},
{"role": "assistant", "content": "$esc.java(${qResult.getText()})" },
#else
{"role": "user", "content": "Give me the $vectaraIdxWord[$foreach.index] search result."},
{"role": "assistant", "content": "$qResult.docMetadata().get('answerDate') $esc.java(${qResult.getText()})" },
#end
#end
{
"role": "user",
"content": "Generate a comprehensive and informative answer for the question \"$esc.java(${vectaraQuery})\" solely
based on the search results in this chat. You must only use information from the provided results. Combine search
results together into a coherent answer. Do not repeat text. Only use the most relevant results that answer the
question accurately. If there are 2 answers that seem in conflict, use the most recent answer according to the
date. If a result does not answer the question, do not use it. If the search results are not valid,
respond with \"The returned results did not contain sufficient information to the question.\"."}
]