chore: optimize (#8)

Author: nduyhai

internal/adapters/http/handler.go

@@ -3,6 +3,8 @@ package http
 import (
 	"log/slog"
 	"net/http"
+	"regexp"
+	"strings"
 
 	tgbotapi "github.com/go-telegram-bot-api/telegram-bot-api/v5"
 	"github.com/nduyhai/valjean/internal/infra/config"
@@ -43,7 +45,7 @@ func (h *Handler) WebHook(c *gin.Context) {
 		MessageID:    upd.Message.MessageID,
 		UserID:       upd.Message.From.ID,
 		UserHandle:   upd.Message.From.UserName,
-		Text:         upd.Message.Text,
+		Text:         h.cleanMention(upd.Message.Text, h.config.Telegram.BotUsername),
 		ContextSnips: h.extractContextSnips(upd),
 		ChatType:     upd.Message.Chat.Type,
 		ReplyFor:     h.getReplyUserName(upd),
@@ -94,3 +96,13 @@ func (h *Handler) extractContextSnips(upd tgbotapi.Update) []string {
 	}
 	return nil
 }
+
+func (h *Handler) cleanMention(text, botUsername string) string {
+	if text == "" || botUsername == "" {
+		return text
+	}
+	re := regexp.MustCompile(`(?i)@` + regexp.QuoteMeta(botUsername) + `\b`)
+	cleaned := re.ReplaceAllString(text, "")
+	cleaned = strings.TrimSpace(strings.Join(strings.Fields(cleaned), " "))
+	return cleaned
+}

internal/adapters/llm/openai/openai.go

@@ -3,6 +3,7 @@ package openai
 import (
 	"context"
 	"log/slog"
+	"strings"
 
 	"github.com/nduyhai/valjean/internal/app/entities"
 	goopenai "github.com/openai/openai-go/v3"
@@ -19,42 +20,82 @@ func NewClient(ai goopenai.Client, logger *slog.Logger) *Client {
 
 func (c *Client) Evaluate(ctx context.Context, in entities.EvalInput) (entities.EvalOutput, error) {
 
-	// Combine context snippets with the main message
-	var fullMessage string
-	if len(in.ContextSnips) > 0 {
-		fullMessage = "Context:\n"
-		for _, contextText := range in.ContextSnips {
-			if contextText != "" {
-				fullMessage += contextText + "\n"
+	// 1) Select & compress context
+	ctxSnips := uniqueNonEmpty(in.ContextSnips)
+	if len(ctxSnips) > 3 {
+		ctxSnips = ctxSnips[len(ctxSnips)-3:] // keep last 3
+	}
+	for i := range ctxSnips {
+		ctxSnips[i] = truncateRunes(strings.TrimSpace(ctxSnips[i]), 240)
+	}
+
+	// 2) Build compact prompt
+	//   ctx lines as "- ..."; then Q: <message>
+	var b strings.Builder
+	if len(ctxSnips) > 0 {
+		for _, s := range ctxSnips {
+			if s != "" {
+				b.WriteString("- ")
+				b.WriteString(s)
+				b.WriteByte('\n')
 			}
 		}
-		fullMessage += "\nUser message:\n" + in.Text
-	} else {
-		fullMessage = in.Text
+		b.WriteByte('\n')
 	}
+	b.WriteString("Q: ")
+	b.WriteString(strings.TrimSpace(in.Text))
+	userPrompt := b.String()
 
+	// 3) Messages: short system + one user
 	messages := []goopenai.ChatCompletionMessageParamUnion{
-		goopenai.UserMessage(fullMessage),
+		goopenai.SystemMessage("You are concise. Answer ≤60 words. If uncertain, say so briefly."),
+		goopenai.UserMessage(userPrompt),
 	}
 
+	// 4) Call with a small model & caps
 	chatCompletion, err := c.ai.Chat.Completions.New(ctx, goopenai.ChatCompletionNewParams{
-		Messages: messages,
-		Model:    goopenai.ChatModelGPT4o,
+		Messages:    messages,
+		Model:       goopenai.ChatModelGPT4oMini, // cheaper than 4o
+		MaxTokens:   goopenai.Int(120),           // cap output tokens
+		Temperature: goopenai.Float(0.2),
 	})
 
 	msg := "Please try again"
 	if err != nil {
-		return entities.EvalOutput{
-			Summary: msg,
-		}, err
+		return entities.EvalOutput{Summary: msg}, err
 	}
-
-	if length := len(chatCompletion.Choices); length > 0 {
-		msg = chatCompletion.Choices[0].Message.Content
+	if len(chatCompletion.Choices) > 0 && chatCompletion.Choices[0].Message.Content != "" {
+		msg = strings.TrimSpace(chatCompletion.Choices[0].Message.Content)
 	}
 	return entities.EvalOutput{
 		Summary:    msg,
-		Citations:  []string{"Citation"},
+		Citations:  nil,
 		Confidence: 80,
 	}, nil
 }
+
+func uniqueNonEmpty(ss []string) []string {
+	seen := make(map[string]struct{}, len(ss))
+	out := make([]string, 0, len(ss))
+	for _, s := range ss {
+		s = strings.TrimSpace(s)
+		if s == "" {
+			continue
+		}
+		key := strings.ToLower(s) // or hash/fingerprint
+		if _, ok := seen[key]; ok {
+			continue
+		}
+		seen[key] = struct{}{}
+		out = append(out, s)
+	}
+	return out
+}
+
+func truncateRunes(s string, max int) string {
+	rs := []rune(s)
+	if len(rs) <= max {
+		return s
+	}
+	return string(rs[:max]) + "…"
+}