audita/internal/modules/glossary/prompt.go

package glossary

import (
	"encoding/json"
	"fmt"

	"gitea.maximumdirect.net/eric/audita/internal/core/schema"
	"gitea.maximumdirect.net/eric/audita/internal/framework/contracts"
	"gitea.maximumdirect.net/eric/audita/internal/framework/promptcontext"
)

type promptSegment struct {
	ID         int      `json:"id"`
	Speaker    string   `json:"speaker"`
	Start      float64  `json:"start"`
	End        float64  `json:"end"`
	Text       string   `json:"text"`
	Categories []string `json:"categories,omitempty"`
}

type promptTranscriptSection struct {
	SectionIndex int             `json:"section_index"`
	Segments     []promptSegment `json:"segments"`
}

func BuildProposalMessages(transcript *schema.Transcript, glossary *schema.Glossary, sectionIndex int, transcriptDescription string) ([]contracts.LLMMessage, error) {
	glossaryJSON, err := json.MarshalIndent(glossary, "", "  ")
	if err != nil {
		return nil, fmt.Errorf("marshal glossary prompt context: %w", err)
	}

	sectionPayload := promptTranscriptSection{
		SectionIndex: sectionIndex,
		Segments:     make([]promptSegment, 0),
	}
	if transcript != nil {
		for _, s := range transcript.Segments {
			sectionPayload.Segments = append(sectionPayload.Segments, promptSegment{
				ID:         s.ID,
				Speaker:    s.Speaker,
				Start:      s.Start,
				End:        s.End,
				Text:       s.Text,
				Categories: append([]string(nil), s.Categories...),
			})
		}
	}
	sectionJSON, err := json.MarshalIndent(sectionPayload, "", "  ")
	if err != nil {
		return nil, fmt.Errorf("marshal transcript prompt context: %w", err)
	}

	system := "You are Audita, a careful transcript correction assistant. Identify only transcription errors that are strongly supported by the glossary. A valid correction must be acoustically plausible: the original transcript text should sound similar to the proposed correction when spoken aloud. Do not make generic grammar, spelling, capitalization, style, or filler-word edits. Do not substitute an unrelated glossary term just because it could fit the topic. Preserve speaker names, timestamps, and meaning."
	user := "Review this transcript section and return only glossary-supported corrections that should be applied.\n\n" +
		"Rules:\n" +
		"- Correct domain-specific names, aliases, jargon, deities, locations, NPCs, players, factions, and similar terms only when both the glossary and surrounding transcript context support the correction.\n" +
		"- The correction must plausibly fix a transcription error: the original words should be phonetically or acoustically similar to the corrected words in spoken English.\n" +
		"- Appropriate example: correcting \"gestures\" to \"Jesters\" can be valid if \"Jesters\" appears in the glossary and nearby context supports that inference.\n" +
		"- Inappropriate example: correcting \"Lyra\" to \"Jesters\" should be omitted because those words are not similar in spoken English, even if \"Jesters\" appears in the glossary.\n" +
		"- Do not replace one clear glossary term, character name, location, or ordinary word with a different glossary term unless it is a plausible mishearing.\n" +
		"- Treat glossary names and aliases already present in the transcript as protected spellings.\n" +
		"- Do not replace, Anglicize, normalize, lowercase, or otherwise alter protected glossary names or aliases away from their glossary spelling.\n" +
		"- Preserve canonical glossary capitalization for protected names and aliases, even if they look unusual.\n" +
		"- If a segment includes categories, treat them as additional transcript context.\n" +
		"- Plural forms of glossary names and aliases are allowed targets when spoken similarity and context support them, even if the plural is not explicitly listed in the glossary.\n" +
		"- Use the exact id from the input segment.\n" +
		"- For returned corrections, original_text must be only the exact text span that needs replacement, not the full segment text unless the whole segment is the replacement span.\n" +
		"- corrected_text must be only the replacement text for that span, not the full corrected segment text unless the whole segment is the replacement span.\n" +
		"- Each returned correction must contain only id, original_text, corrected_text, and confidence.\n" +
		"- Do not return corrections where original_text and corrected_text are identical.\n" +
		"- Do not return speaker, start, or end fields.\n" +
		"- Return only changed segments; do not return entries for unchanged segments.\n" +
		"- confidence must be between 0.0 and 1.0.\n" +
		"- If no corrections are needed, return an empty corrections list.\n\n" +
		promptcontext.TranscriptDescriptionBlock(transcriptDescription) +
		fmt.Sprintf("Glossary:\n%s\n\nTranscript section:\n%s", string(glossaryJSON), string(sectionJSON))

	return []contracts.LLMMessage{
		{Role: "system", Content: system},
		{Role: "user", Content: user},
	}, nil
}