Lifeline AI

title
Analysis Example #1
Analysis Example #2
Analysis Example #3
Analysis Example #4
Analysis Example #5
Analysis Example #6
PDF Generation Example #1
PDF Generation Example #2
PDF Generation Example #3
Mail Example

Inspiration

I built Lifeline AI from personal experience, I was in toxic relationships without even realizing it. Looking back, I was completely oblivious to the manipulation patterns and red flags that seem obvious now. I'd question my own memories when told "that never happened," accept blame for things that weren't my fault, and rationalize concerning behaviors as normal.

What I needed was an objective observer, something or someone that could analyze interactions without the emotional blindness that kept me trapped. That's when I realized AI could be that impartial witness having access to a myriad of specialized books in relationships to base its judge on, thus the processing of evidence would be done without the gaslighting, emotional manipulation, or self-doubt that clouds human judgment.

What it does

Lifeline AI serves as your digital advocate, processing relationship evidence through multiple sophisticated layers:

Evidence Ingestion: Upload digital evidence, like screenshots of conversations, audio recordings, or videos. The AWS Lambda pipeline automatically processes everything, transcribing audio/video via AWS Transcribe and extracting text from images using AWS Rekognition.

AI Analysis: Using gpt-oss-120b through Groq, analyze relationship dynamics against established psychological frameworks (Gottman's Four Horsemen, attachment theory, trauma patterns). The system generates:

Toxicity scores (0-100) with confidence intervals
Attachment style assessments
Communication pattern analysis
Risk level indicators

Professional Documentation: Through Foxit API integration, there are generated PDF reports that offer clinical insights for therapists, simplified explanations for family, evidence summaries for lawyers, each tailored to maximize understanding and action.

How I built it

Backend Architecture:

Spring Boot (Java 17) REST API with JWT authentication
TiDB Cloud for distributed MySQL-compatible database
FastAPI (Python) service for AI processing

Cloud Infrastructure:

5 AWS Lambda functions for serverless evidence processing
AWS EventBridge for event-driven orchestration
AWS S3 with lifecycle policies for secure file storage
AWS DynamoDB for real-time processing status tracking

AI Integration:

Groq API with gpt-oss-120b for relationship analysis
Foxit PDF API for automated document workflow
AWS Transcribe for audio/video transcription
AWS Rekognition for image text extraction

Frontend:

React 19 with TypeScript for type-safe UI
EmailJS for secure communication with trusted contacts
Vite for optimized build pipeline

The entire system processes evidence through an event-driven pipeline: Upload → S3 → EventBridge → Lambda Processing → AI Analysis → PDF Generation → Distribution.

Challenges I ran into

AI Prompt Engineering: Creating prompts that could accurately detect subtle manipulation tactics while avoiding false positives took numerous iterations. I had to balance sensitivity with accuracy, especially for gaslighting detection.

PDF Generation Pipeline: Implementing a reliable HTML-to-PDF conversion with Foxit API that maintained professional formatting while handling dynamic content required custom styling and multiple retry mechanisms.

Accomplishments that we're proud of

Complete automation from evidence upload to professional PDF generation in under 60 seconds
Role-adaptive communication that successfully tailors messages for different recipients (therapists, lawyers, family)

Most importantly, I built something that would have helped me recognize and document my own toxic relationships years earlier.

What I learned

EventBridge rules are powerful for orchestrating complex serverless workflows
Groq's implementation of gpt-oss-120b provides excellent performance for sensitive content analysis
Building this forced me to confront my own past experiences analytically

What's next for Lifeline AI

Real-time monitoring integration with messaging apps (with consent)
Therapist portal for professional case management
AI fine-tuning on specialized datasets for cultural context awareness

My vision is to make Lifeline AI the standard tool for relationship safety—available to anyone who needs an objective view of their situation, before it's too late.