Soul Scanner

The AI bodyguard for your mind. Decodes gaslighting & power dynamics. A hyper-vertical AI mining truth for neurodivergent safety.

Comment

💡 Inspiration: The "Black Box" of Social Interaction

For the 15-20% of the population who are Neurodivergent (Autism, ADHD, CPTSD), social interaction is often a minefield. We are high-performance Ferraris forced to drive on muddy roads.

While we possess high cognitive processing power, we often lack the "decryption key" for neurotypical subtext.

  • Is this person gaslighting me?
  • Is this a passive-aggressive attack?
  • What is the hidden power dynamic here?

The cost of missing these cues is burnout, victimization, and trauma. We realized that existing AI assistants are too "polite" or "neutral." We didn't need a therapist to ask "how does that make you feel?" We needed a bodyguard to tell us "this is a trap, and here is how you dismantle it."

🛡️ What it does

Soul Scanner is an AI-powered Cognitive Prosthesis. It acts as a forensic text analyst that sits between the user and the chaotic social world.

It doesn't just read text; it reads intent. Using a specialized "Three-Mind Architecture," it analyzes incoming messages to:

  1. Mask X-Ray: Instantly detect manipulation techniques like Gaslighting, Triangulation, and Projection.
  2. Power Dynamics Analysis: Calculate the "DPI" (Dynamic Power Index) of the conversation to see who holds the leverage.
  3. Tactical Defense: Generate "Grey Rock" responses—boring, non-defensive replies designed to de-escalate conflict and protect the user's energy.

⚙️ How we built it (The "Soul" in the Machine)

We built Soul Scanner using Google Gemini 1.5 Pro as the core reasoning engine.

  • The Kernel: We injected a curated knowledge base of 17 classic psychology texts (including The Body Keeps the Score and The Art of War) into the System Prompt. This grounds the AI's analysis in established behavioral science, not just probability.
  • Prompt Engineering: We utilized "Chain of Thought" prompting to force Gemini to act as a cold, objective analyst. It must classify the logical fallacies before generating advice.
  • The Stack: The frontend is built with Next.js and React, deployed on Vercel for instant, edge-network speed. We used the Vercel AI SDK to stream Gemini's responses in real-time, reducing the latency between "attack" and "defense."

🧩 Challenges we ran into

  • Balancing Tone: Standard LLMs are RLHF-trained to be helpful and agreeable. Our challenge was "jailbreaking" that politeness to make the AI sharp, direct, and protective—like an INTJ strategist.
  • Reducing Hallucinations: We had to strictly constrain the AI to only cite psychological concepts that exist in its provided kernel, ensuring users receive accurate terminology for their reality testing.

🏆 Accomplishments that we're proud of

  • The "Mask X-Ray" Accuracy: The model can now successfully identify subtle passive-aggressive phrasing ("I'm just joking," "You're too sensitive") that usually slips past standard content filters.
  • Real-Time Latency: Achieving a near-instant analysis speed that allows users to use it during a stressful text exchange, not just after.

🧠 What we learned

We learned that Gemini's reasoning capabilities are exceptionally suited for "Subtext Analysis." Unlike other models that focus on surface-level sentiment (Positive/Negative), Gemini proved capable of understanding complex interpersonal context and power asymmetry.

🚀 What's next for Soul Scanner

  • Browser Extension: A "Cognitive Firewall" that overlays analysis directly onto Gmail, WhatsApp, and Slack.
  • Voice Mode: Using Gemini's multimodal capabilities to analyze tone of voice in real-time calls.
  • The Mission: To democratize "Social Safety" for the neurodivergent community, turning vulnerability into a tactical advantage.

Built With

  • engineering
  • google-gemini-api
  • javascript
  • next.js
  • prompt
  • react
  • vercel
  • vercel-ai-sdk
Share this project:

Updates