Thrive Church AWS Lambdas

AWS Lambda functions for Thrive Church automation, including podcast RSS feed generation, sermon processing, and AI-powered content enrichment.

Architecture

┌─────────────────────────────────────────────────────────────────────────────┐
│                           SERMON PROCESSING PIPELINE                        │
└─────────────────────────────────────────────────────────────────────────────┘

┌──────────────────┐     ┌──────────────────────┐     ┌────────────────────┐
│   .NET API       │     │  transcription_      │     │  sermon_processor  │
│  (App Runner)    │────▶│  processor           │───▶│                    │
│                  │     │                      │     │  • Summary (GPT)   │
│ PodcastLambda    │     │  • Download audio    │     │  • Tags (GPT)      │
│ Service.cs       │     │  • Azure Speech API  │     │  • Waveform        │
│                  │     │  • Store transcript  │     │  • Update MongoDB  │
└──────────────────┘     └──────────┬───────────┘     │    Messages        │
                                    │                 └─────────┬──────────┘
                                    │                           │
                                    │                           │ (if series has EndDate)
                                    │                           ▼
                                    │                 ┌────────────────────┐
                                    │                 │  series_summary_   │
                                    │                 │  processor         │
                                    │                 │                    │
                                    │                 │  • Series Summary  │
                                    │                 │    (GPT)           │
                                    │                 │  • Update MongoDB  │
                                    │                 │    Series          │
                                    │                 └────────────────────┘
                                    │
                                    │                 ┌────────────────────┐
                                    └────────────────▶│  podcast_rss_      │
                                                      │  generator         │
                                                      │                    │
                                                      │  • Description(GPT)│
                                                      │  • Upsert MongoDB  │
                                                      │    PodcastEpisodes │
                                                      │  • Update RSS XML  │
                                                      │    to S3           │
                                                      └────────────────────┘

What Gets Generated

Content Type	Where	Description
Message Summary	MongoDB Messages.Summary	TLDR-style, 130-180 words, uses we/us/our perspective, leads with the topic/lesson
Topical Tags	MongoDB Messages.Tags	90+ tags across 17 categories (Theological, Spiritual Disciplines, Personal Growth, etc.)
Waveform Data	MongoDB Messages.WaveformData	200-point waveform for audio player visualization
Series Summary	MongoDB Series.Summary	Present-tense timeless truths about the series themes (generated when series is complete)
Podcast Description	MongoDB PodcastEpisodes.Description, S3 RSS XML	Two paragraphs (130-180 words) for non-church audience, no speaker names

Prompt Engineering

All prompts have been refined through systematic evaluation using promptfoo. Key prompt characteristics:

• Message Summaries: TLDR-style, straightforward educational tone, first person plural (we/us/our)
• Series Summaries: Present-tense timeless truths, neutral contemplative tone, varied openings
• Podcast Descriptions: Two paragraphs for spiritual seekers, names specific tensions, ends with curiosity (not questions)
• Tags: Comprehensive categorization using predefined tag taxonomy mapped to C# enums

Series Summary Flow

When a message is processed by sermon_processor, it checks if the message's series has an EndDate (indicating the series is complete). If so, it triggers series_summary_processor which:

1. Verifies all messages with audio in the series have summaries (race condition protection)
2. Aggregates all message summaries
3. Generates a cohesive series-level summary using GPT
4. Saves the summary to the Series document in MongoDB

Lambdas

Lambda	Purpose
transcription-processor	Downloads audio from S3, transcribes with Azure Speech API, stores transcript in Azure Blob, invokes sermon-processor and podcast-rss-generator
sermon-processor	Generates message summary (TLDR-style), tags (90+), waveform. Triggers series-summary-processor for completed series
podcast-rss-generator	Generates podcast descriptions (two paragraphs, no speaker names), updates PodcastEpisodes and RSS XML
series-summary-processor	Generates series-level summaries as present-tense timeless truths

Project Structure

AWSLambdas/
├── README.md
├── template.yaml                 # SAM template
├── requirements-dev.txt          # Development dependencies
├── lambdas/
│   ├── podcast_rss_generator/
│   │   ├── handler.py
│   │   ├── prompts/              # Prompt files
│   │   └── requirements.txt
│   ├── sermon_processor/
│   │   ├── handler.py
│   │   ├── prompts/              # Prompt files
│   │   └── requirements.txt
│   ├── series_summary_processor/
│   │   ├── handler.py
│   │   ├── prompts/              # Prompt files
│   │   └── requirements.txt
│   ├── transcription_processor/
│   │   ├── handler.py
│   │   └── requirements.txt
│   └── shared/                   # Shared utilities
│       └── ...
├── promptfoo/                    # Prompt evaluation configs and tests
│   ├── config_*.yaml             # Evaluation configurations
│   ├── prompts/                  # Prompt templates for evaluation
│   ├── tests/                    # Dynamic test generators
│   └── golden_examples/          # Reference examples
├── scripts/                      # Backfill and utility scripts
│   └── ...
├── tests/
│   └── ...
└── docs/
    └── ...

Local Development

Prerequisites

• Python 3.11+
• AWS CLI configured
• MongoDB connection string
• Langfuse (for LLM visibility)

Setup

python -m venv venv
venv\Scripts\activate  # Windows
pip install -r requirements-dev.txt

Environment Variables

Environment variables are configured in template.yaml and injected at runtime. Secrets are stored in AWS Secrets Manager.

Transcription Processor

Variable	Description
S3_BUCKET	S3 bucket for audio files (e.g., thrive-audio)
SERMON_LAMBDA_NAME	Name of sermon processor Lambda to invoke
PODCAST_LAMBDA_NAME	Name of podcast RSS generator Lambda to invoke
AZURE_STORAGE_ACCOUNT	Azure Blob Storage account for transcripts
AZURE_STORAGE_CONTAINER	Azure container name (e.g., transcripts)
AZURE_SPEECH_ENDPOINT	Azure Speech-to-Text endpoint

Sermon Processor

Variable	Description
S3_BUCKET	S3 bucket for audio files
SERIES_SUMMARY_LAMBDA_NAME	Name of series summary Lambda to invoke

Series Summary Processor

Variable	Description
(uses Secrets Manager)	MongoDB URI and OpenAI API key from Secrets Manager

Podcast RSS Generator

Variable	Description
RSS_BUCKET	S3 bucket for RSS XML file
RSS_KEY	S3 key for RSS XML file

Deployment

Lambdas are deployed using AWS SAM via GitHub Actions CI/CD.

# Validate template
sam validate

# Build all Lambdas
sam build

# Deploy (requires AWS credentials)
sam deploy --guided

See template.yaml for full configuration including IAM roles, log groups, and Lambda layers.

Related Projects

• ThriveChurchOfficialAPI - Main C# API
• ThriveAPIMediaTool - Admin Tool UI