feat: asking AI for a description and storing vectors on DB

backend/agents/client/embeddings.go

@@ -0,0 +1,102 @@
+package client
+
+import (
+	"bytes"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"os"
+)
+
+type EmbeddingsClient struct {
+	url    string
+	apiKey string
+
+	Do func(req *http.Request) (*http.Response, error)
+}
+
+type EmbeddingsRequest struct {
+	Model string `json:"model"`
+	Input string `json:"input"`
+}
+
+type EmbeddingsData struct {
+	Object     string    `json:"object"`
+	Index      int       `json:"index"`
+	Embeddings []float64 `json:"embedding"`
+}
+
+type EmbeddingsResponse struct {
+	Object string           `json:"object"`
+	Data   []EmbeddingsData `json:"data"`
+}
+
+func CreateEmbeddingsClient() EmbeddingsClient {
+	apiKey := os.Getenv(OPENAI_API_KEY)
+
+	if len(apiKey) == 0 {
+		panic("No api key")
+	}
+
+	return EmbeddingsClient{
+		apiKey: apiKey,
+		url:    "https://api.openai.com/v1/embeddings",
+		Do: func(req *http.Request) (*http.Response, error) {
+			client := &http.Client{}
+			return client.Do(req)
+		},
+	}
+}
+
+func (client EmbeddingsClient) getRequest(text string) (*http.Request, error) {
+	embeddingsReq := EmbeddingsRequest{
+		Model: "text-embedding-3-large",
+		Input: text,
+	}
+
+	jsonEmbeddingsBody, err := json.Marshal(embeddingsReq)
+	if err != nil {
+		return nil, err
+	}
+
+	req, err := http.NewRequest("POST", client.url, bytes.NewBuffer(jsonEmbeddingsBody))
+	if err != nil {
+		return req, err
+	}
+
+	req.Header.Add("Authorization", "Bearer "+client.apiKey)
+	req.Header.Add("Content-Type", "application/json")
+
+	return req, nil
+}
+
+func (client EmbeddingsClient) Request(text string) (EmbeddingsData, error) {
+	httpRequest, err := client.getRequest(text)
+	if err != nil {
+		return EmbeddingsData{}, err
+	}
+
+	resp, err := client.Do(httpRequest)
+	if err != nil {
+		return EmbeddingsData{}, err
+	}
+
+	response, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return EmbeddingsData{}, err
+	}
+
+	embeddingsResponse := EmbeddingsResponse{}
+	err = json.Unmarshal(response, &embeddingsResponse)
+
+	if err != nil {
+		return EmbeddingsData{}, err
+	}
+
+	if len(embeddingsResponse.Data) != 1 {
+		return EmbeddingsData{}, errors.New("Unsupported. We currently only accept 1 choice from AI.")
+	}
+
+	return embeddingsResponse.Data[0], nil
+}

backend/agents/embeddings_agent.go

@@ -0,0 +1,68 @@
+package agents
+
+import (
+	"context"
+	"screenmark/screenmark/agents/client"
+	"screenmark/screenmark/models"
+
+	"github.com/charmbracelet/log"
+	"github.com/google/uuid"
+)
+
+const embeddingsAgentPropmt = `
+You are an agent who's job it is to describe the contents of an image.
+This description should be detailed as it will be used to create embeddings from the image.
+You should focus more on the content of the image, rather than it's appearence.
+`
+
+type EmbeddingAgent struct {
+	embeddings client.EmbeddingsClient
+
+	client client.AgentClient
+
+	imageModel models.ImageModel
+
+	log *log.Logger
+}
+
+func (agent EmbeddingAgent) GetEmbeddings(userId uuid.UUID, imageId uuid.UUID, imageName string, imageData []byte) error {
+	request := client.AgentRequestBody{
+		Model:       "gpt-4.1-mini",
+		Temperature: 0.3,
+		ResponseFormat: client.ResponseFormat{
+			Type: "text",
+		},
+		Chat: &client.Chat{
+			Messages: make([]client.ChatMessage, 0),
+		},
+	}
+
+	request.Chat.AddSystem(embeddingsAgentPropmt)
+	request.Chat.AddImage(imageName, imageData, nil)
+
+	resp, err := agent.client.Request(&request)
+	if err != nil {
+		return err
+	}
+
+	description := resp.Choices[0].Message.Content
+	log.Info(description)
+
+	embeddings, err := agent.embeddings.Request(description)
+	if err != nil {
+		return err
+	}
+
+	return agent.imageModel.UpdateEmbedding(context.Background(), imageId, embeddings.Embeddings)
+}
+
+func NewEmbeddingsAgent(log *log.Logger, imageModel models.ImageModel) EmbeddingAgent {
+	return EmbeddingAgent{
+		client: client.CreateAgentClient(client.CreateAgentClientOptions{
+			SystemPrompt: embeddingsAgentPropmt,
+			Log:          log,
+		}),
+		embeddings: client.CreateEmbeddingsClient(),
+		imageModel: imageModel,
+	}
+}

backend/events.go

@@ -67,11 +67,19 @@ func ListenNewImageEvents(db *sql.DB, notifier *Notifier[Notification]) {
 				locationAgent := agents.NewLocationAgent(createLogger("Locations 📍", splitWriter), locationModel)
 				eventAgent := agents.NewEventAgent(createLogger("Events 📅", splitWriter), eventModel, locationModel)
 
+				embeddings := agents.NewEmbeddingsAgent(createLogger("Embeddings 📊", splitWriter), imageModel)
+
 				if err := imageModel.StartProcessing(ctx, image.ID); err != nil {
 					databaseEventLog.Error("Failed to FinishProcessing", "error", err)
 					return
 				}
 
+				err = embeddings.GetEmbeddings(image.UserID, image.ImageID, image.Image.ImageName, image.Image.Image)
+				if err != nil {
+					databaseEventLog.Error("Failed to get embeddings", "error", err)
+					return
+				}
+
 				orchestrator := agents.NewOrchestratorAgent(createLogger("Orchestrator 🎼", splitWriter), noteAgent, contactAgent, locationAgent, eventAgent, image.Image.ImageName, image.Image.Image)
 				orchestrator.RunAgent(image.UserID, image.ImageID, image.Image.ImageName, image.Image.Image)
 				_, err = imageModel.FinishProcessing(ctx, image.ID)
@@ -184,7 +192,6 @@ func CreateEventsHandler(notifier *Notifier[Notification]) http.HandlerFunc {
 					return
 				}
 
-				fmt.Printf("Sending msg %s\n", msg)
 				fmt.Fprintf(w, "event: data\ndata: %s\n\n", string(msgString))
 				w.(http.Flusher).Flush()
 			}

backend/models/image.go

@@ -7,6 +7,8 @@ import (
 	"fmt"
 	"screenmark/screenmark/.gen/haystack/haystack/model"
 	. "screenmark/screenmark/.gen/haystack/haystack/table"
+	"strconv"
+	"strings"
 
 	. "github.com/go-jet/jet/v2/postgres"
 
@@ -180,6 +182,20 @@ func (m ImageModel) IsUserAuthorized(ctx context.Context, imageId uuid.UUID, use
 	return err != nil && userImage.UserID.String() == userId.String()
 }
 
+func (m ImageModel) UpdateEmbedding(ctx context.Context, imageId uuid.UUID, embedding []float64) error {
+	stringSlice := make([]string, len(embedding))
+
+	for i, f := range embedding {
+		stringSlice[i] = strconv.FormatFloat(f, 'f', -1, 64)
+	}
+
+	embeddingString := "[" + strings.Join(stringSlice, ",") + "]"
+
+	_, err := m.dbPool.ExecContext(ctx, "UPDATE haystack.image SET embedding = $1::vector WHERE id = $2", embeddingString, imageId)
+
+	return err
+}
+
 func NewImageModel(db *sql.DB) ImageModel {
 	return ImageModel{dbPool: db}
 }

backend/schema.sql

@@ -1,6 +1,7 @@
 DROP SCHEMA IF EXISTS haystack CASCADE;
 
 CREATE SCHEMA haystack;
+CREATE EXTENSION IF NOT EXISTS vector;
 
 /* -----| Enums |----- */
 
@@ -16,7 +17,9 @@ CREATE TABLE haystack.users (
 CREATE TABLE haystack.image (
 	id uuid PRIMARY KEY DEFAULT gen_random_uuid(),
     image_name TEXT NOT NULL,
-	image BYTEA NOT NULL
+	image BYTEA NOT NULL,
+
+    embedding vector(3072)
 );
 
 CREATE TABLE haystack.user_images_to_process (