kunthawat/ALwrity
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
f2103101776b08a483b0188d7e49c88d1ef2f635
ALwrity/backend/api/podcast/handlers/analysis.py

580 lines
24 KiB
Python
Raw Blame History

"""
Podcast Analysis Handlers
Analysis endpoint for podcast ideas.
"""
from fastapi import APIRouter, Depends, HTTPException
from typing import Dict, Any, Optional, List
import json
import uuid
from sqlalchemy.orm import Session
from pydantic import BaseModel
from services.database import get_db
from middleware.auth_middleware import get_current_user
from api.story_writer.utils.auth import require_authenticated_user
from services.llm_providers.main_text_generation import llm_text_gen
from services.llm_providers.main_image_generation import generate_image
from services.podcast_bible_service import PodcastBibleService
from services.subscription import PricingService
from models.subscription_models import APIProvider
from utils.asset_tracker import save_asset_to_library
from loguru import logger
import os
from ..constants import PODCAST_IMAGES_DIR
from ..models import (
PodcastAnalyzeRequest,
PodcastAnalyzeResponse,
PodcastEnhanceIdeaRequest,
PodcastEnhanceIdeaResponse
)
# Check if running in podcast-only demo mode
def _is_podcast_only_mode() -> bool:
"""Check if podcast-only demo mode is enabled."""
return os.getenv("ALWRITY_ENABLED_FEATURES", "").strip().lower() == "podcast"
def _estimate_tokens(text: str) -> int:
if not text:
return 0
return max(1, len(text) // 4)
def _build_analysis_estimate(
db: Session,
idea: str,
duration: int,
speakers: int,
has_avatar: bool,
) -> Dict[str, Any]:
"""
Build a user-facing estimate from pricing catalog and phase-level assumptions.
"""
# Defaults if catalog lookup fails
gemini_in_token = 0.00000015
gemini_out_token = 0.0000006
exa_per_request = 0.005
image_per_request = 0.01
video_per_request = 0.01
audio_per_request = 0.005
try:
pricing_service = PricingService(db)
gemini_pricing = pricing_service.get_pricing_for_provider_model(APIProvider.GEMINI, "gemini-2.5-flash") or {}
gemini_in_token = float(gemini_pricing.get("cost_per_input_token") or gemini_in_token)
gemini_out_token = float(gemini_pricing.get("cost_per_output_token") or gemini_out_token)
exa_pricing = pricing_service.get_pricing_for_provider_model(APIProvider.EXA, "exa-search") or {}
exa_per_request = float(exa_pricing.get("cost_per_request") or exa_per_request)
img_pricing = pricing_service.get_pricing_for_provider_model(APIProvider.STABILITY, "stable-image-ultra") or {}
image_per_request = float(img_pricing.get("cost_per_request") or image_per_request)
video_pricing = pricing_service.get_pricing_for_provider_model(APIProvider.VIDEO, "minimax-video-01") or {}
video_per_request = float(video_pricing.get("cost_per_request") or video_per_request)
audio_pricing = pricing_service.get_pricing_for_provider_model(APIProvider.AUDIO, "gemini-2.5-flash-preview-tts") or {}
audio_per_request = float(audio_pricing.get("cost_per_request") or audio_per_request)
except Exception as exc:
logger.warning(f"[Podcast Analyze] Pricing catalog lookup failed, using defaults: {exc}")
# Phase assumptions
query_count = 5
analyze_in = _estimate_tokens(idea) + 240
analyze_out = 750
analyze_cost = (analyze_in * gemini_in_token) + (analyze_out * gemini_out_token)
gather_cost = query_count * exa_per_request
script_chars = max(1000, duration * 900)
write_in = _estimate_tokens(idea) + _estimate_tokens(str(script_chars)) + 320
write_out = max(900, int(duration * 220))
write_cost = (write_in * gemini_in_token) + (write_out * gemini_out_token)
tts_cost = max(1, speakers) * audio_per_request
avatar_cost = 0.0 if has_avatar else image_per_request
video_cost = max(1, duration) * video_per_request
produce_cost = tts_cost + avatar_cost + video_cost
breakdown = [
{"phase": "Analyze", "cost": round(analyze_cost, 6)},
{"phase": "Gather", "cost": round(gather_cost, 6)},
{"phase": "Write", "cost": round(write_cost, 6)},
{"phase": "Produce", "cost": round(produce_cost, 6)},
]
total = round(sum(item["cost"] for item in breakdown), 6)
return {
"ttsCost": round(tts_cost, 6),
"avatarCost": round(avatar_cost, 6),
"videoCost": round(video_cost, 6),
"researchCost": round(gather_cost, 6),
"total": total,
"breakdown": breakdown,
"currency": "USD",
}
router = APIRouter()
@router.post("/idea/enhance", response_model=PodcastEnhanceIdeaResponse)
async def enhance_podcast_idea(
request: PodcastEnhanceIdeaRequest,
current_user: Dict[str, Any] = Depends(get_current_user),
):
"""
Take raw keywords/topic and use AI to craft a presentable, detailed podcast idea.
Uses the user's Podcast Bible for hyper-personalization if available.
"""
user_id = require_authenticated_user(current_user)
# Serialize Bible context if provided or generate from onboarding
# In podcast-only mode, skip bible generation since onboarding is disabled
bible_context = ""
if not _is_podcast_only_mode():
logger.warning(f"[Podcast Enhance] Podcast mode=full — attempting Bible generation for user {user_id}")
try:
bible_service = PodcastBibleService()
if request.bible:
from models.podcast_bible_models import PodcastBible
bible_data = PodcastBible(**request.bible)
bible_context = bible_service.serialize_bible(bible_data)
else:
# Generate from onboarding data directly
bible_obj = bible_service.generate_bible(user_id, "temp_enhance")
bible_context = bible_service.serialize_bible(bible_obj)
except Exception as exc:
logger.warning(f"[Podcast Enhance] Failed to parse or generate bible context: {exc}")
else:
# In podcast mode, use the provided bible directly if available
logger.warning(f"[Podcast Enhance] Podcast mode=podcast_only — skipping Bible generation for user {user_id}")
if request.bible:
try:
from models.podcast_bible_models import PodcastBible
bible_data = PodcastBible(**request.bible)
bible_service = PodcastBibleService()
bible_context = bible_service.serialize_bible(bible_data)
except Exception as exc:
logger.debug(f"[Podcast Enhance] Bible parsing skipped in podcast mode: {exc}")
prompt = f"""
You are a creative podcast producer. Generate 3 distinct, compelling podcast episode concepts from the raw idea.
{f"USER PERSONALIZATION CONTEXT (Podcast Bible):\n{bible_context}\n" if bible_context else ""}
RAW IDEA/KEYWORDS: "{request.idea}"
TASK:
Generate 3 different enhanced versions, each with a unique angle:
1. Professional & Expert-led angle (focus on authority, insights, and expertise)
2. Storytelling & Human interest angle (focus on narratives, emotions, and personal connections)
3. Trendy & Contemporary angle (focus on current trends, modern perspectives, and relevance)
Each version should be 2-3 sentences, audience-focused, and align with host persona if provided.
Return JSON with:
- enhanced_ideas: array of 3 strings, each string being a complete episode pitch (NOT objects, just plain strings)
- rationales: array of 3 strings explaining the approach for each version
IMPORTANT: enhanced_ideas must be an array of plain strings, NOT objects. Example:
{{
"enhanced_ideas": [
"Your expert guide to AI advancement: A practical look at how AI is transforming industries...",
"The human stories behind AI innovation: From Silicon Valley to your daily life...",
"AI in 2026: What's trending and what's next in artificial intelligence..."
],
"rationales": [
"Professional approach focusing on expertise and authority",
"Storytelling approach emphasizing human connection",
"Contemporary approach highlighting current relevance"
]
}}
"""
try:
raw = llm_text_gen(
prompt=prompt,
user_id=user_id,
json_struct=None,
preferred_provider=None,
flow_type="premium_tool",
)
# Normalize response
if isinstance(raw, str):
data = json.loads(raw)
else:
data = raw
# Extract enhanced ideas and rationales with fallbacks
enhanced_ideas = data.get("enhanced_ideas", [])
rationales = data.get("rationales", [])
# Handle case where LLM returns objects instead of strings
normalized_ideas = []
for idea in enhanced_ideas:
if isinstance(idea, dict):
# Extract title and description from object
title = idea.get("title", "")
description = idea.get("description", "") or idea.get("content", "")
normalized_ideas.append(f"{title}: {description}" if description else title)
elif isinstance(idea, str):
normalized_ideas.append(idea)
enhanced_ideas = normalized_ideas
# Ensure we have exactly 3 ideas, fallback to original if needed
if not isinstance(enhanced_ideas, list) or len(enhanced_ideas) != 3:
# Fallback: create 3 variations of the original idea
base_idea = request.idea
enhanced_ideas = [
f"Expert insights on {base_idea}: A deep dive into industry trends and best practices.",
f"The human side of {base_idea}: Personal stories and real-world experiences that resonate.",
f"Modern perspectives on {base_idea}: Current trends and forward-thinking approaches."
]
rationales = [
"Professional approach focusing on expertise and authority",
"Storytelling approach emphasizing human connection",
"Contemporary approach highlighting current relevance"
]
# Ensure rationales match the number of ideas
if not isinstance(rationales, list) or len(rationales) != 3:
rationales = [
"Professional angle with expert insights",
"Storytelling angle with human interest",
"Trendy angle with contemporary relevance"
]
return PodcastEnhanceIdeaResponse(
enhanced_ideas=enhanced_ideas[:3], # Ensure exactly 3
rationales=rationales[:3] # Ensure exactly 3
)
except HTTPException:
# Re-raise HTTPExceptions (e.g., 429 subscription limit) - preserve error details
raise
except Exception as exc:
logger.error(f"[Podcast Enhance] Failed for user {user_id}: {exc}")
raise HTTPException(status_code=500, detail=f"Enhance failed: {exc}")
@router.post("/analyze", response_model=PodcastAnalyzeResponse)
async def analyze_podcast_idea(
request: PodcastAnalyzeRequest,
current_user: Dict[str, Any] = Depends(get_current_user),
db: Session = Depends(get_db),
):
"""
Analyze a podcast idea and return podcast-oriented outlines, keywords, and titles.
If no avatar_url is provided, it generates one automatically based on the host's look.
"""
user_id = require_authenticated_user(current_user)
# Serialize Bible context if provided or generate from onboarding
bible_context = ""
bible_obj = None
try:
bible_service = PodcastBibleService()
if request.bible:
from models.podcast_bible_models import PodcastBible
bible_data = PodcastBible(**request.bible)
bible_context = bible_service.serialize_bible(bible_data)
bible_obj = bible_data
else:
# Generate from onboarding data directly
bible_obj = bible_service.generate_bible(user_id, "temp_analyze")
bible_context = bible_service.serialize_bible(bible_obj)
bible_obj = bible_obj
except Exception as exc:
logger.warning(f"[Podcast Analyze] Failed to parse or generate bible context: {exc}")
# --- NEW: Generate Presenter Avatar if missing ---
final_avatar_url = request.avatar_url
final_avatar_prompt = None
# Skip avatar generation for audio_only mode
podcast_mode = getattr(request, 'podcast_mode', None) or 'video_only'
should_generate_avatar = not final_avatar_url and podcast_mode != 'audio_only'
if should_generate_avatar:
logger.info(f"[Podcast Analyze] No avatar_url provided, generating one for user {user_id}")
try:
# 1. PRE-FLIGHT VALIDATION: Check subscription limits for image generation
from services.subscription import PricingService
from services.subscription.preflight_validator import validate_image_generation_operations
pricing_service = PricingService(db)
validate_image_generation_operations(
pricing_service=pricing_service,
user_id=user_id,
num_images=1
)
# 2. Build avatar prompt from Bible host look or fallback
host_look = bible_obj.host.look if bible_obj and bible_obj.host.look else "A professional podcast host"
visual_style = bible_obj.visual_style.style_preset if bible_obj else "Realistic Photography"
final_avatar_prompt = f"Professional headshot of a podcast host, {host_look}, {visual_style} style, clean background, soft studio lighting, center-focused, high resolution, sharp focus, professional photography quality, 16:9 aspect ratio."
# 3. Generate the image
logger.info(f"[Podcast Analyze] Generating avatar with prompt: {final_avatar_prompt}")
image_result = generate_image(
prompt=final_avatar_prompt,
user_id=user_id,
options={"width": 1024, "height": 1024}
)
# 4. Save to disk and library
if image_result and image_result.image_bytes:
img_id = str(uuid.uuid4())[:8]
filename = f"presenter_podcast_{user_id}_{img_id}.png"
avatars_dir = PODCAST_IMAGES_DIR / "avatars"
avatars_dir.mkdir(parents=True, exist_ok=True)
output_path = avatars_dir / filename
with open(output_path, "wb") as f:
f.write(image_result.image_bytes)
final_avatar_url = f"/api/podcast/images/avatars/{filename}"
# Save to asset library for reuse
save_asset_to_library(
db=db,
user_id=user_id,
asset_type="image",
source_module="podcast_analysis",
filename=filename,
file_url=final_avatar_url,
title=f"Presenter Avatar - {request.idea[:40]}",
description=f"AI-generated podcast presenter for: {request.idea}",
provider=image_result.provider,
model=image_result.model,
cost=0.0 # Cost tracked in generate_image
)
logger.info(f"[Podcast Analyze] ✅ Generated and saved avatar to {final_avatar_url}")
except Exception as e:
logger.error(f"[Podcast Analyze] ❌ Failed to generate avatar: {e}")
# Non-fatal: continue analysis even if avatar generation fails
# --- END: Avatar Generation ---
# Incorporate user feedback if provided
feedback_context = ""
if request.feedback:
feedback_context = f"""
USER REGENERATION FEEDBACK:
The user was not satisfied with the previous analysis. They provided the following instructions for improvement:
"{request.feedback}"
Please prioritize this feedback and adjust the analysis accordingly.
"""
prompt = f"""
You are an expert podcast producer and research strategist. Given a podcast idea, craft concise podcast-ready assets
that sound like episode plans (not fiction stories).
{f"USER PERSONALIZATION CONTEXT (Podcast Bible):\n{bible_context}\n" if bible_context else ""}
{feedback_context}
Podcast Idea: "{request.idea}"
Duration: ~{request.duration} minutes
Speakers: {request.speakers} (host + optional guest)
TASK:
1. Define the target audience and content type aligned with the Bible's "Audience DNA" and "Brand DNA".
2. Identify 5 high-impact keywords.
3. Propose 2 episode outlines with factual segments.
4. Suggest 3 titles.
5. IMPORTANT: Generate 4-6 specific research queries for Exa. These queries MUST be highly targeted to the episode's topic, the host's expertise level, and the audience's interests as defined in the Bible.
* Do NOT use generic queries like "latest trends in X".
* DO use queries that look for case studies, specific data points, expert opinions, or contrasting viewpoints that would make for a deep, insightful podcast conversation.
Return JSON with:
- audience: short target audience description
- content_type: podcast style/format
- top_keywords: 5 podcast-relevant keywords/phrases
- suggested_outlines: 2 items, each with title (<=60 chars) and 4-6 short segments (bullet-friendly, factual)
- title_suggestions: 3 concise episode titles
- episode_hook: one compelling 15-30 second opening hook/angle that grabs attention
- key_takeaways: 3-5 actionable insights listeners will learn
- guest_talking_points: (if guest included) 3-4 suggested questions/angles for guest interview
- listener_cta: one clear call-to-action for listeners
- research_queries: array of {{"query": "string", "rationale": "string"}}
- exa_suggested_config: suggested Exa search options with:
- exa_search_type: "auto" | "neural" | "keyword"
- exa_category: one of ["research paper","news","company","github","tweet","personal site","pdf","financial report","linkedin profile"]
- exa_include_domains: up to 3 reputable domains
- exa_exclude_domains: up to 3 domains
- max_sources: 6-10
- include_statistics: boolean
- date_range: one of ["last_month","last_3_months","last_year","all_time"]
Requirements:
- Keep language factual, actionable, and suited for spoken audio.
- Avoid narrative fiction tone.
- For research queries: Mix of time-sensitive and evergreen queries:
- 2-3 queries should focus on latest 2025-2026 developments, trends, and data (use year in query)
- 2-3 queries should be evergreen/fundamental (concepts, definitions, best practices, proven strategies) - do NOT include years in these
- Today's date is April 2026.
"""
try:
raw = llm_text_gen(
prompt=prompt,
user_id=user_id,
json_struct=None,
preferred_provider=None,
flow_type="premium_tool",
)
except HTTPException:
# Re-raise HTTPExceptions (e.g., 429 subscription limit) - preserve error details
raise
except Exception as exc:
logger.error(f"[Podcast Analyze] Analysis failed for user {user_id}: {exc}")
raise HTTPException(status_code=500, detail=f"Analysis failed: {exc}")
# Normalize response (accept dict or JSON string)
if isinstance(raw, str):
try:
data = json.loads(raw)
except json.JSONDecodeError:
raise HTTPException(status_code=500, detail="LLM returned non-JSON output")
elif isinstance(raw, dict):
data = raw
else:
raise HTTPException(status_code=500, detail="Unexpected LLM response format")
audience = data.get("audience") or "Growth-focused professionals"
content_type = data.get("content_type") or "Interview + insights"
top_keywords = data.get("top_keywords") or []
suggested_outlines = data.get("suggested_outlines") or []
title_suggestions = data.get("title_suggestions") or []
episode_hook = data.get("episode_hook") or ""
key_takeaways = data.get("key_takeaways") or []
guest_talking_points = data.get("guest_talking_points") or []
listener_cta = data.get("listener_cta") or ""
research_queries = data.get("research_queries") or []
exa_suggested_config = data.get("exa_suggested_config") or None
return PodcastAnalyzeResponse(
audience=audience,
content_type=content_type,
top_keywords=top_keywords,
suggested_outlines=suggested_outlines,
title_suggestions=title_suggestions,
episode_hook=episode_hook,
key_takeaways=key_takeaways,
guest_talking_points=guest_talking_points,
listener_cta=listener_cta,
research_queries=research_queries,
exa_suggested_config=exa_suggested_config,
bible=bible_obj.model_dump() if bible_obj else None,
avatar_url=final_avatar_url,
avatar_prompt=final_avatar_prompt,
estimate=_build_analysis_estimate(
db=db,
idea=request.idea,
duration=request.duration,
speakers=request.speakers,
has_avatar=bool(final_avatar_url),
),
)
class RegenerateQueriesRequest(BaseModel):
idea: str
feedback: str
existing_analysis: Optional[Dict[str, Any]] = None
bible: Optional[Dict[str, Any]] = None
class RegenerateQueriesResponse(BaseModel):
research_queries: List[Dict[str, str]]
@router.post("/regenerate-queries", response_model=RegenerateQueriesResponse)
async def regenerate_research_queries(
request: RegenerateQueriesRequest,
current_user: Dict[str, Any] = Depends(get_current_user),
):
"""
Regenerate research queries based on user feedback and existing analysis.
"""
user_id = require_authenticated_user(current_user)
# Build context from existing analysis
idea = request.idea
feedback = request.feedback
# Get topic, keywords, audience from existing analysis if provided
topic = idea
keywords = ""
audience = ""
if request.existing_analysis:
topic = request.existing_analysis.get("title_suggestions", [idea])[0] if request.existing_analysis.get("title_suggestions") else idea
keywords = ", ".join(request.existing_analysis.get("top_keywords", [])[:5])
audience = request.existing_analysis.get("audience", "")
# Serialize Bible context if provided
bible_context = ""
if request.bible:
try:
bible_service = PodcastBibleService()
from models.podcast_bible_models import PodcastBible
bible_data = PodcastBible(**request.bible)
bible_context = bible_service.serialize_bible(bible_data)
except Exception as e:
logger.warning(f"Failed to serialize bible for query regeneration: {e}")
prompt = f"""
You are a research strategist for podcast content. Given a podcast idea, existing analysis, and user feedback,
generate 7 new research queries that address the user's specific needs.
{f"USER FEEDBACK: {feedback}" if feedback else ""}
{f"EXISTING ANALYSIS CONTEXT:\n- Topic: {topic}\n- Keywords: {keywords}\n- Audience: {audience}\n" if request.existing_analysis else ""}
{f"PODCAST BIBLE CONTEXT:\n{bible_context}\n" if bible_context else ""}
Podcast Idea: "{idea}"
TASK:
Generate exactly 7 research queries that:
1. Incorporate the user's feedback direction
2. Build on the existing analysis context
3. Mix of time-sensitive (2025-2026) and evergreen topics
4. Are highly specific to the podcast topic
Return JSON with:
- research_queries: array of {{"query": "string", "rationale": "string"}}
Requirements:
- At least 2-3 queries should focus on latest 2025-2026 developments (include year in query)
- At least 2-3 queries should be evergreen (concepts, definitions, best practices - NO year)
- Queries should be specific and actionable, not generic
"""
try:
from services.llm_providers.main_text_generation import llm_text_gen
raw = llm_text_gen(
prompt=prompt,
user_id=user_id,
json_struct={"research_queries": [{"query": "string", "rationale": "string"}]},
preferred_provider=None,
flow_type="premium_tool",
)
# Parse response
if isinstance(raw, dict):
queries = raw.get("research_queries", [])
else:
# Try to parse as JSON
try:
parsed = json.loads(raw) if isinstance(raw, str) else raw
queries = parsed.get("research_queries", []) if isinstance(parsed, dict) else []
except:
queries = []
return RegenerateQueriesResponse(research_queries=queries[:7])
except HTTPException:
raise
except Exception as exc:
logger.error(f"[Regenerate Queries] Failed for user {user_id}: {exc}")
raise HTTPException(status_code=500, detail=f"Regenerate queries failed: {exc}")
Reference in New Issue View Git Blame Copy Permalink