video rag example (#63)

* rag for video

* upgrade video from rag

* add complete app

* add deployment instructions

.gitignore

@@ -167,5 +167,6 @@ cython_debug/
 #  and can be added to the global gitignore or merged into this file.  For a more nuclear
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
 #.idea/
+
 *.json
 .DS_Store

examples/docker/video-rag/Dockerfile

@@ -0,0 +1,10 @@
+FROM python:3.11
+
+COPY app.py app.py
+COPY .env .env
+COPY requirements.txt requirements.txt
+RUN pip install torch==2.1.1 torchvision==0.16.1 --index-url https://download.pytorch.org/whl/cpu
+COPY static/ static/
+RUN pip install -r requirements.txt
+
+ENTRYPOINT ["solara", "run", "app.py", "--host=0.0.0.0", "--port=80"]

examples/docker/video-rag/README.md

@@ -0,0 +1,73 @@
+## RAG from video
+
+This is a full stack project that uses a video as input, populates an ElasticSearch instance with vectors through an indexing pipeline, and initializes a retriver pipeline with a Solara app.
+
+A complete walk through of this application can be found [here](https://ploomber.io/blog/rag-video-app/)
+
+### Requirements
+
+- Python 3.11
+- Docker
+- OpenAI Key
+
+### Installation
+
+1. Clone the repository
+2. Create a virtual environment with with Python 3.11 and activate it
+3. Install the requirements through the `requirements.txt` file
+
+### If you are running a local ElasticSearch instance
+
+Use docker-compose to run the ElasticSearch instance
+
+```bash
+docker-compose up -d
+```
+
+### If you are running a remote ElasticSearch instance
+
+Create a `.env` file with the following variables:
+
+```bash
+elastic_search_host=<host>
+elastic_username=<username>
+elastic_password=<password>
+OPENAI=<key>
+```
+
+Replace host, username, and password with the corresponding values for your cloud-based instance.
+
+Modify the document store in `video_indexing.py` and `app.py` to use the remote instance.
+
+### Execution of indexing pipeline
+
+```bash
+python indexing_pipeline.py
+```
+
+### Local execution of retriever pipeline through the Solara app
+
+```bash
+cd app/
+solara run app.py
+```
+
+### Deploy app on Ploomber cloud (assumes cloud-based set up of an ElasticSearch instance)
+
+Generate an API key on [Ploomber Cloud](https://www.platform.ploomber.io/applications) under Account. Paste your API key via the CLI. 
+
+```bash
+ploomber-cloud key
+```
+
+Initialize your deployment environment
+
+```bash
+ploomber-cloud init
+```
+
+Deploy your app
+
+```bash
+ploomber-cloud deploy
+```

examples/docker/video-rag/app.py

@@ -0,0 +1,212 @@
+from pathlib import Path
+from dataclasses import dataclass
+
+import solara
+
+from haystack import Pipeline
+from haystack.components.embedders import SentenceTransformersTextEmbedder
+from haystack.components.builders.prompt_builder import PromptBuilder
+from haystack.components.generators import GPTGenerator
+from elasticsearch_haystack.embedding_retriever import ElasticsearchEmbeddingRetriever
+from elasticsearch_haystack.document_store import ElasticsearchDocumentStore
+from dotenv import load_dotenv
+import os
+
+load_dotenv(".env")
+openaikey = os.getenv("OPENAI")
+elastic_search_cloud_id = os.getenv("elastic_search_cloud_id")
+elastic_search_host = os.getenv("elastic_search_host")
+elastic_username = os.getenv("elastic_username")
+elastic_password = os.getenv("elastic_password")
+#
+# Build RAG pipeline
+print("Initializing QA pipeline")
+prompt_template = """\
+Use the following context to answer the user's question in a friendly manner. \
+    If the context provided doesn't answer the question - \
+        please respond with: "There is no information in my knowledge base about this".
+
+### CONTEXT
+{% for doc in documents %}
+  {{ doc.content }}
+{% endfor %}
+
+### USER QUESTION
+{{query}}
+"""
+
+#document_store = ElasticsearchDocumentStore(hosts= "http://localhost:9200/")
+document_store = ElasticsearchDocumentStore(hosts=elastic_search_host,
+                                            basic_auth=(elastic_username, elastic_password))
+
+prompt_builder = PromptBuilder(prompt_template)
+############################################
+query_embedder = SentenceTransformersTextEmbedder()
+retriever = ElasticsearchEmbeddingRetriever(document_store=document_store)
+llm = GPTGenerator(api_key=openaikey)
+
+pipeline = Pipeline()
+pipeline.add_component(instance=query_embedder, name="query_embedder")
+pipeline.add_component(instance=retriever, name="retriever")
+pipeline.add_component(instance=prompt_builder, name="prompt_builder")
+pipeline.add_component(instance=llm, name="llm")
+
+pipeline.connect("query_embedder.embedding", "retriever.query_embedding")
+pipeline.connect("retriever.documents", "prompt_builder.documents")
+pipeline.connect("prompt_builder", "llm")
+
+###########################################
+# Solara app
+
+class State:
+    input = solara.reactive("")
+
+css = """
+    .main {
+        width: 100%;
+        height: 100%;
+        max-width: 1200px;
+        margin: auto;
+        padding: 1em;
+    }
+    
+    #app > div > div:nth-child(2) > div:nth-child(2) {
+    display: none;
+}
+"""
+
+chatbox_css = """
+.message {
+    max-width: 450px;
+    width: 100%;
+}
+
+.user-message, .user-message > * {
+    background-color: #f0f0f0 !important;
+}
+
+.assistant-message, .assistant-message > * {
+    background-color: #9ab2e9 !important;
+}
+
+.avatar {
+  width: 50px;
+  height: 50px;
+  border-radius: 50%;
+  border: 2px solid transparent;
+  overflow: hidden;
+  display: flex;
+}
+
+.avatar img {
+  width: 100%;
+  height: 100%;
+  object-fit: cover;
+}
+"""
+
+
+@dataclass
+class Message:
+    role: str
+    content: str
+
+
+def ChatBox(message: Message) -> None:
+    solara.Style(chatbox_css)
+
+    align = "start" if message.role == "assistant" else "end"
+    with solara.Column(align=align):
+        with solara.Card(classes=["message", f"{message.role}-message"]):
+            if message.content:
+                with solara.Card():
+                    solara.Markdown(message.content)
+
+
+        # Image reference: https://www.flaticon.com/free-icons/bot;
+        #                  https://www.flaticon.com/free-icons/use
+
+        with solara.HBox(align_items="center"):
+            image_path = Path(f"static/{message.role}-logo.png")
+            solara.Image(str(image_path), classes=["avatar"])
+            solara.Text(message.role.capitalize())
+
+@solara.component
+def Chat() -> None:
+    solara.Style(
+        """
+        .chat-input {
+            max-width: 800px;
+        })
+    """
+    )
+
+    messages, set_messages = solara.use_state(
+        [
+            Message(
+                role="assistant",
+                content=f"Welcome. Please post your queries! My knowledge base\
+                    has been curated on a small collection of videos from NASA.  \
+                    This collection of videos consist of short clips that talk \
+                    about the topics: Mars Perseverance Rover.\
+                    Sample questions: \n\nWhat is the Mars Perseverance Rover? \
+                        What is the Mars Perseverance Rover mission? \
+                        Tell me about the helicopter on Mars."
+            )
+        ]
+    )
+    input, set_input = solara.use_state("")
+
+    def ask_rag(pipeline):
+        try:
+            input_text = State.input.value
+            _messages = messages + [Message(role="user", content=input_text)]
+            set_input("")
+            State.input.value = ""
+            set_messages(_messages)
+
+            result = pipeline.run(data={"query_embedder": {"text": input_text}, "prompt_builder": {"query": input_text}})
+            rag_response = result['llm']['replies'][0]
+
+            set_messages(_messages + [Message(role="assistant", content=rag_response)])
+
+        except Exception as e:
+            set_messages(_messages + [Message(role="assistant", content=f"Cannot answer your current question. Please try again {e}")])
+
+    with solara.VBox():
+        for message in messages:
+            ChatBox(message)
+
+    with solara.Row(justify="center"):
+        with solara.HBox(align_items="center", classes=["chat-input"]):
+            solara.InputText(label="Query", value=State.input, continuous_update=False)
+
+    if State.input.value:
+        ask_rag(pipeline)
+
+@solara.component
+def Page():
+
+    with solara.AppBarTitle():
+        solara.Text("Deepen your understanding of our video collection through a Q&A AI assistant")
+
+    with solara.Card(title="About", elevation=6, style="background-color: #f5f5f5;"):
+        with solara.Row(justify="center"):
+            solara.Image(image="static/nasa-logo.svg", width="100")  # Adjust width and height as needed
+
+        solara.Markdown("Ask questions about our curated database of video using advanced AI tools. \n \
+                        This database is curated from the following list of videos: \n \
+                        https://images.nasa.gov/search?q=nasa%20perseverance%20rover&page=1&media=video&yearStart=2023&yearEnd=2024")
+
+    solara.Style(css)
+    with solara.VBox(classes=["main"]):
+        solara.HTML(
+            tag="h3", style="margin: auto;", unsafe_innerHTML="Chat with the assistant to answer questions about the video topics"
+        )
+
+        Chat()
+
+@solara.component
+def Layout(children):
+    route, routes = solara.use_route()
+    return solara.AppLayout(children=children)

examples/docker/video-rag/docker-compose.yml

@@ -0,0 +1,15 @@
+services:
+  elasticsearch:
+    image: "docker.elastic.co/elasticsearch/elasticsearch:8.11.1"
+    ports:
+      - 9200:9200
+    restart: on-failure
+    environment:
+      - discovery.type=single-node
+      - xpack.security.enabled=false
+      - "ES_JAVA_OPTS=-Xms1024m -Xmx1024m"
+    healthcheck:
+        test: curl --fail http://localhost:9200/_cat/health || exit 1
+        interval: 10s
+        timeout: 1s
+        retries: 10

examples/docker/video-rag/requirements.txt

@@ -0,0 +1,8 @@
+haystack-ai==2.0.0b3
+elasticsearch-haystack==0.1.2
+solara
+python-dotenv
+sentence-transformers>=2.2.0
+moviepy 
+pydub
+ploomber-cloud