Update to new PromptingTools RAG (#10)

Artifacts.toml

@@ -5,3 +5,51 @@ lazy = true
     [[juliaextra.download]]
     sha256 = "61133afa7e06fda133f07164c57190a5b922f8f2a1aa17c3f8a628b5cf752512"
     url = "https://github.com/svilupp/AIHelpMeArtifacts/raw/main/artifacts/juliaextra__v1.10.0__ada1.0.tar.gz"
+
+["julia__textembedding3large-0-Float32"]
+git-tree-sha1 = "a105a2482296fa0a80ce0c76677cc9ef673be70e"
+lazy = true
+
+    [["julia__textembedding3large-0-Float32".download]]
+    sha256 = "ff4e91908fb54b7919aad9d6a2ac5045124d43eb864fe9f96a7a68d304d4e0a2"
+    url = "https://github.com/svilupp/AIHelpMeArtifacts/raw/main/artifacts/julia__v1.10.2__textembedding3large-0-Float32__v1.0.tar.gz"
+
+["julia__textembedding3large-1024-Bool"]
+git-tree-sha1 = "7eef82f15c72712b4f5fff2449ebf3ed64b56b14"
+lazy = true
+
+    [["julia__textembedding3large-1024-Bool".download]]
+    sha256 = "27186886d19ea4c3f1710b4bc70e8e809d906069d5de8c992c948d97d0f454da"
+    url = "https://github.com/svilupp/AIHelpMeArtifacts/raw/main/artifacts/julia__v1.10.2__textembedding3large-1024-Bool__v1.0.tar.gz"
+
+["tidier__textembedding3large-0-Float32"]
+git-tree-sha1 = "680c7035e512844fd2b9af1757b02b931dfadaa5"
+lazy = true
+
+    [["tidier__textembedding3large-0-Float32".download]]
+    sha256 = "59eb6fef198e32d238c11d3a95e5201d18cb83c5d42eae753706614c0f72db9e"
+    url = "https://github.com/svilupp/AIHelpMeArtifacts/raw/main/artifacts/tidier__v20240407__textembedding3large-0-Float32__v1.0.tar.gz"
+
+["tidier__textembedding3large-1024-Bool"]
+git-tree-sha1 = "44d861977d663a9c4615023ae38828e0ef88036e"
+lazy = true
+
+    [["tidier__textembedding3large-1024-Bool".download]]
+    sha256 = "226cadd2805abb6ab6e561330aca97466e0a2cb1e1eb171be661d9dea9dcacdc"
+    url = "https://github.com/svilupp/AIHelpMeArtifacts/raw/main/artifacts/tidier__v20240407__textembedding3large-1024-Bool__v1.0.tar.gz"
+
+["makie__textembedding3large-0-Float32"]
+git-tree-sha1 = "30c29c10d9b2b160b43f358fad9f4f6fe83ce378"
+lazy = true
+
+    [["makie__textembedding3large-0-Float32".download]]
+    sha256 = "ee15489022df191fbede93adf1bd7cc1ceb1f84185229026a5e38ae9a3fab737"
+    url = "https://github.com/svilupp/AIHelpMeArtifacts/raw/main/artifacts/makie__v20240330__textembedding3large-0-Float32__v1.0.tar.gz"
+
+["makie__textembedding3large-1024-Bool"]
+git-tree-sha1 = "a49a86949f86f6cf4c29bdc9559c05064b49c801"
+lazy = true
+
+    [["makie__textembedding3large-1024-Bool".download]]
+    sha256 = "135f36effc0d29ed20e9bc877f727e4d9d8366bcae4bf4d13f998529d1091324"
+    url = "https://github.com/svilupp/AIHelpMeArtifacts/raw/main/artifacts/makie__v20240330__textembedding3large-1024-Bool__v1.0.tar.gz"

CHANGELOG.md

@@ -7,6 +7,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Added
+- (Preliminary) Knowledge packs available for Julia docs (`:julia`), Tidier ecosystem (`:tidier`), Makie ecosystem (`:makie`). Load with `load_index!(:julia)` or several with `load_index!([:julia, :tidier])`.
+
+### Changed
+- Bumped up PromptingTools to v0.20 (brings new RAG capabilities, pretty-printing, etc.)
+- Changed default model to be GPT-4 Turbo to improve answer quality
 
 ### Fixed
 - Fixed wrong initiation of `CONV_HISTORY` and other globals that led to UndefVarError. Moved several globals to `const Ref{}` pattern to ensure type stability, but it means that from now it always needs to be dereferenced with `[]` (eg, `MAIN_INDEX[]` instead of `MAIN_INDEX`).

Project.toml

@@ -4,8 +4,11 @@ authors = ["J S <[email protected]> and contributors"]
 version = "0.0.1-DEV"
 
 [deps]
+HDF5 = "f67ccb44-e63f-5c2f-98bd-6dc0ccc4ba2f"
 LazyArtifacts = "4af54fe1-eca0-43a8-85a7-787d91b784e3"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+Logging = "56ddb016-857b-54e1-b83d-db4d58db5568"
+PrecompileTools = "aea7be01-6a6a-4083-8856-8a6e6704d82a"
 Preferences = "21216c6a-2e73-6563-6e65-726566657250"
 PromptingTools = "670122d1-24a8-4d70-bfce-740807c42192"
 REPL = "3fa0cd96-eef1-5676-8a61-b3b8758bbffb"
@@ -20,7 +23,7 @@ JSON3 = "1"
 LazyArtifacts = "<0.0.1, 1"
 LinearAlgebra = "<0.0.1, 1"
 Preferences = "1"
-PromptingTools = "0.9"
+PromptingTools = "0.20"
 REPL = "1"
 SHA = "0.7"
 Serialization = "<0.0.1, 1"

README.md

@@ -4,7 +4,8 @@
 
 AIHelpMe harnesses the power of Julia's extensive documentation and advanced AI models to provide tailored coding guidance. By integrating with PromptingTools.jl, it offers a unique, AI-assisted approach to answering your coding queries directly in Julia's environment.
 
-Note: This is only a proof-of-concept. If there is enough interest, we will fine-tune the RAG pipeline for better performance.
+> [!CAUTION]
+> This is only a proof-of-concept. If there is enough interest, we will fine-tune the RAG pipeline for better performance.
 
 ## Features
 
@@ -27,7 +28,8 @@ Pkg.add(url="https://github.com/svilupp/AIHelpMe.jl")
 
 - Julia (version 1.10 or later).
 - Internet connection for API access.
-- OpenAI and Cohere API keys (recommended for optimal performance). See [How to Obtain API Keys](#how-to-obtain-api-keys).
+- OpenAI API keys with available credits. See [How to Obtain API Keys](#how-to-obtain-api-keys).
+- For optimal performance, get also Cohere API key (free for community use) and Tavily API key (free for community use).
 
 All setup should take less than 5 minutes!
 
@@ -40,10 +42,43 @@ All setup should take less than 5 minutes!
     ```
 
     ```plaintext
-   [ Info: Done generating response. Total cost: $0.001
+   [ Info: Done generating response. Total cost: $0.015
    AIMessage("To implement quicksort in Julia, you can use the `sort` function with the `alg=QuickSort` argument.")
    ```
 
+   Note: As a default, we load only the Julia documentation and docstrings for standard libraries. The default model used is GPT-4 Turbo.
+
+   You can pretty-print the answer using `pprint` if you return the full RAGResult (`return_all=true`):
+   ```julia
+   using AIHelpMe: pprint
+
+   result = aihelp("How do I implement quicksort in Julia?", return_all=true)
+   pprint(result)
+   ```
+
+   ```plaintext
+   --------------------
+   QUESTION(s)
+   --------------------
+   - How do I implement quicksort in Julia?
+
+   --------------------
+   ANSWER
+   --------------------
+   To implement quicksort in Julia, you can use the [5,1.0]`sort`[1,1.0] function with the [1,1.0]`alg=QuickSort`[1,1.0] argument.[2,1.0]
+
+   --------------------
+   SOURCES
+   --------------------
+   1. https://docs.julialang.org/en/v1.10.2/base/sort/index.html::Sorting and Related Functions/Sorting Functions
+   2. https://docs.julialang.org/en/v1.10.2/base/sort/index.html::Sorting and Related Functions/Sorting Functions
+   3. https://docs.julialang.org/en/v1.10.2/base/sort/index.html::Sorting and Related Functions/Sorting Algorithms
+   4. SortingAlgorithms::/README.md::0::SortingAlgorithms
+   5. AIHelpMe::/README.md::0::AIHelpMe
+   ```
+
+   Note: You can see the model cheated because it can see this very documentation...
+
 2. **`aihelp` Macro**:
    ```julia
    aihelp"how to implement quicksort in Julia?"
@@ -56,11 +91,12 @@ All setup should take less than 5 minutes!
    Note: The `!` is required for follow-up questions.
    `aihelp!` does not add new context/more information - to do that, you need to ask a new question.
 
-4. **Pick stronger models**:
-    Eg, "gpt4t" is an alias for GPT-4 Turbo:
+4. **Pick faster models**:
+    Eg, for simple questions, GPT 3.5 might be enough, so use the alias "gpt3t":
     ```julia
-    aihelp"Elaborate on the `sort` function and quicksort algorithm"gpt4t
+    aihelp"Elaborate on the `sort` function and quicksort algorithm"gpt3t
     ```
+
     ```plaintext
     [ Info: Done generating response. Total cost: $0.002 -->
     AIMessage("The `sort` function in programming languages, including Julia.... continues for a while!
@@ -69,22 +105,36 @@ All setup should take less than 5 minutes!
 5. **Debugging**:
    How did you come up with that answer? Check the "context" provided to the AI model (ie, the documentation snippets that were used to generate the answer):
     ```julia
-    const AHM = AIHelpMe
-    AHM.preview_context()
+    AIHelpMe.pprint(AIHelpMe.LAST_RESULT[])
     # Output: Pretty-printed Question + Context + Answer with color highlights
     ```
 
+    The color highlights show you which words were NOT supported by the provided context (magenta = completely new, blue = partially new). 
+    It's a quite and intuitive way to see which function names or variables are made up versus which ones were in the context. 
+
+    You can change the kwargs of `pprint` to hide the annotations or potentially even show the underlying context (snippets from the documentation):
+
+    ```julia
+    AIHelpMe.pprint(AIHelpMe.LAST_RESULT[]; add_context = true, add_scores = false)
+    ```
+
 ## How to Obtain API Keys
 
 ### OpenAI API Key:
 1. Visit [OpenAI's API portal](https://openai.com/api/).
 2. Sign up and generate an API Key.
-3. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://svilupp.github.io/PromptingTools.jl/dev/frequently_asked_questions/#Configuring-the-Environment-Variable-for-API-Key).
+3. Charge some credits ($5 minimum but that will last you for a lost time).
+4. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://siml.earth/PromptingTools.jl/dev/frequently_asked_questions#Configuring-the-Environment-Variable-for-API-Key).
 
 ### Cohere API Key:
 1. Sign up at [Cohere's registration page](https://dashboard.cohere.com/welcome/register).
 2. After registering, visit the [API keys section](https://dashboard.cohere.com/api-keys) to obtain a free, rate-limited Trial key.
-3. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://svilupp.github.io/PromptingTools.jl/dev/frequently_asked_questions/#Configuring-the-Environment-Variable-for-API-Key).
+3. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://siml.earth/PromptingTools.jl/dev/frequently_asked_questions#Configuring-the-Environment-Variable-for-API-Key).
+
+### Tavily API Key:
+1. Sign up at [Tavily](https://app.tavily.com/sign-in).
+2. After registering, generate an API key on the [Overview page](https://app.tavily.com/home). You can get a free, rate-limited Trial key.
+3. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://siml.earth/PromptingTools.jl/dev/frequently_asked_questions#Configuring-the-Environment-Variable-for-API-Key).
 
 ## Usage
 

docs/src/index.md

@@ -24,14 +24,15 @@ To install AIHelpMe, use the Julia package manager and the address of the reposi
 
 ```julia
 using Pkg
-Pkg.add("https://github.com/svilupp/AIHelpMe.jl")
+Pkg.add(url="https://github.com/svilupp/AIHelpMe.jl")
 ```
 
 **Prerequisites:**
 
 - Julia (version 1.10 or later).
 - Internet connection for API access.
-- OpenAI and Cohere API keys (recommended for optimal performance). See [How to Obtain API Keys](#how-to-obtain-api-keys).
+- OpenAI API keys with available credits. See [How to Obtain API Keys](#how-to-obtain-api-keys).
+- For optimal performance, get also Cohere API key (free for community use) and Tavily API key (free for community use).
 
 All setup should take less than 5 minutes!
 
@@ -44,10 +45,43 @@ All setup should take less than 5 minutes!
     ```
 
     ```plaintext
-   [ Info: Done generating response. Total cost: $0.001
+   [ Info: Done generating response. Total cost: $0.015
    AIMessage("To implement quicksort in Julia, you can use the `sort` function with the `alg=QuickSort` argument.")
    ```
 
+   Note: As a default, we load only the Julia documentation and docstrings for standard libraries. The default model used is GPT-4 Turbo.
+
+   You can pretty-print the answer using `pprint` if you return the full RAGResult (`return_all=true`):
+   ```julia
+   using AIHelpMe: pprint
+
+   result = aihelp("How do I implement quicksort in Julia?", return_all=true)
+   pprint(result)
+   ```
+
+   ```plaintext
+   --------------------
+   QUESTION(s)
+   --------------------
+   - How do I implement quicksort in Julia?
+
+   --------------------
+   ANSWER
+   --------------------
+   To implement quicksort in Julia, you can use the [5,1.0]`sort`[1,1.0] function with the [1,1.0]`alg=QuickSort`[1,1.0] argument.[2,1.0]
+
+   --------------------
+   SOURCES
+   --------------------
+   1. https://docs.julialang.org/en/v1.10.2/base/sort/index.html::Sorting and Related Functions/Sorting Functions
+   2. https://docs.julialang.org/en/v1.10.2/base/sort/index.html::Sorting and Related Functions/Sorting Functions
+   3. https://docs.julialang.org/en/v1.10.2/base/sort/index.html::Sorting and Related Functions/Sorting Algorithms
+   4. SortingAlgorithms::/README.md::0::SortingAlgorithms
+   5. AIHelpMe::/README.md::0::AIHelpMe
+   ```
+
+   Note: You can see the model cheated because it can see this very documentation...
+
 2. **`aihelp` Macro**:
    ```julia
    aihelp"how to implement quicksort in Julia?"
@@ -60,11 +94,12 @@ All setup should take less than 5 minutes!
    Note: The `!` is required for follow-up questions.
    `aihelp!` does not add new context/more information - to do that, you need to ask a new question.
 
-4. **Pick stronger models**:
-    Eg, "gpt4t" is an alias for GPT-4 Turbo:
+4. **Pick faster models**:
+    Eg, for simple questions, GPT 3.5 might be enough, so use the alias "gpt3t":
     ```julia
-    aihelp"Elaborate on the `sort` function and quicksort algorithm"gpt4t
+    aihelp"Elaborate on the `sort` function and quicksort algorithm"gpt3t
     ```
+
     ```plaintext
     [ Info: Done generating response. Total cost: $0.002 -->
     AIMessage("The `sort` function in programming languages, including Julia.... continues for a while!
@@ -73,22 +108,36 @@ All setup should take less than 5 minutes!
 5. **Debugging**:
    How did you come up with that answer? Check the "context" provided to the AI model (ie, the documentation snippets that were used to generate the answer):
     ```julia
-    const AHM = AIHelpMe
-    AHM.preview_context()
+    AIHelpMe.pprint(AIHelpMe.LAST_RESULT[])
     # Output: Pretty-printed Question + Context + Answer with color highlights
     ```
 
+    The color highlights show you which words were NOT supported by the provided context (magenta = completely new, blue = partially new). 
+    It's a quite and intuitive way to see which function names or variables are made up versus which ones were in the context. 
+
+    You can change the kwargs of `pprint` to hide the annotations or potentially even show the underlying context (snippets from the documentation):
+
+    ```julia
+    AIHelpMe.pprint(AIHelpMe.LAST_RESULT[]; add_context = true, add_scores = false)
+    ```
+
 ## How to Obtain API Keys
 
 ### OpenAI API Key:
 1. Visit [OpenAI's API portal](https://openai.com/api/).
 2. Sign up and generate an API Key.
-3. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://svilupp.github.io/PromptingTools.jl/dev/frequently_asked_questions/#Configuring-the-Environment-Variable-for-API-Key).
+3. Charge some credits ($5 minimum but that will last you for a lost time).
+4. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://siml.earth/PromptingTools.jl/dev/frequently_asked_questions#Configuring-the-Environment-Variable-for-API-Key).
 
 ### Cohere API Key:
 1. Sign up at [Cohere's registration page](https://dashboard.cohere.com/welcome/register).
 2. After registering, visit the [API keys section](https://dashboard.cohere.com/api-keys) to obtain a free, rate-limited Trial key.
-3. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://svilupp.github.io/PromptingTools.jl/dev/frequently_asked_questions/#Configuring-the-Environment-Variable-for-API-Key).
+3. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://siml.earth/PromptingTools.jl/dev/frequently_asked_questions#Configuring-the-Environment-Variable-for-API-Key).
+
+### Tavily API Key:
+1. Sign up at [Tavily](https://app.tavily.com/sign-in).
+2. After registering, generate an API key on the [Overview page](https://app.tavily.com/home). You can get a free, rate-limited Trial key.
+3. Set it as an environment variable or a local preference in PromptingTools.jl. See the [instructions](https://siml.earth/PromptingTools.jl/dev/frequently_asked_questions#Configuring-the-Environment-Variable-for-API-Key).
 
 ## Usage
 

src/AIHelpMe.jl

@@ -4,35 +4,46 @@ using Preferences, Serialization, LinearAlgebra, SparseArrays
 using LazyArtifacts
 using Base.Docs: DocStr, MultiDoc, doc, meta
 using REPL: stripmd
+using HDF5
 
 using PromptingTools
+using PromptingTools: pprint
 using PromptingTools.Experimental.RAGTools
+using PromptingTools.Experimental.RAGTools: AbstractRAGConfig, getpropertynested,
+                                            setpropertynested, merge_kwargs_nested
 using SHA: sha256, bytes2hex
+using Logging, PrecompileTools
 const PT = PromptingTools
-const RAG = PromptingTools.Experimental.RAGTools
+const RT = PromptingTools.Experimental.RAGTools
 
-## export load_index!, last_context, update_index!
 ## export remove_pkgdir, annotate_source, find_new_chunks
 include("utils.jl")
 
+## Globals and types are defined in here
+include("pipeline_defaults.jl")
+
 ## export docdata_to_source, docextract, build_index
 include("preparation.jl")
 
+## export load_index!, update_index!
+include("loading.jl")
+
 export aihelp
 include("generation.jl")
 
 export @aihelp_str, @aihelp!_str
 include("macros.jl")
 
-## Globals
-const CONV_HISTORY = Vector{Vector{PT.AbstractMessage}}()
-const CONV_HISTORY_LOCK = ReentrantLock()
-const MAX_HISTORY_LENGTH = 1
-const LAST_CONTEXT = Ref{Union{Nothing, RAG.RAGContext}}(nothing)
-const MAIN_INDEX = Ref{Union{Nothing, RAG.AbstractChunkIndex}}(nothing)
 function __init__()
-    ## Load index
-    MAIN_INDEX[] = load_index!()
+    ## Set the active configuration
+    update_pipeline!(:bronze)
+    ## Load index - auto-loads into MAIN_INDEX
+    load_index!(:julia)
+end
+
+# Enable precompilation to reduce start time, disabled logging
+with_logger(NullLogger()) do
+    @compile_workload include("precompilation.jl")
 end
 
 end