This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
SAATHI AI: What If We Could Recognize the Warning Signs Before It's Too Late?
**
**AI Detects. Humans Understand. You Decide.
Sometimes, an unsafe situation doesn't begin with an obvious threat.
It can begin with someone constantly asking where you are. Someone getting angry when you spend time with friends. Repeated messages from different accounts. Or a person showing up at places you visit and making you feel watched.
One incident may be difficult to understand on its own. But when these behaviours become patterns, a person may start feeling uncomfortable, controlled, or unsafe.
I started building SAATHI AI around one question: Can AI help people recognize concerning behaviour earlier, understand possible warning signs, and explore safer next steps without taking away their choices?
That question became the starting point for my project.
💜 What Is SAATHI AI?
SAATHI AI is a women's safety and early-intervention prototype designed to help people recognize potential warning signs of controlling behaviour, harassment, stalking, and intimidation.
The name SAATHI represents the idea of having a companion—a tool that helps someone think through a difficult situation without judging them or making decisions for them.
The goal is not to make AI act like a police officer, judge, or therapist.
The goal is to make safety awareness more approachable.
Recognize Early. Reach Help. Protect Dignity.
🚨 The Problem I Wanted to Explore
When someone experiences controlling behaviour or repeated unwanted contact, they may not immediately know how to interpret what is happening.
They might ask themselves:
Am I overthinking this?
Is this behaviour becoming a pattern?
What can I do without making the situation worse?
Where can I look for reliable support?
A website cannot answer every question, and AI cannot know the complete truth about someone's situation.
However, a carefully designed tool may help someone organize their thoughts, recognize possible warning signs, and explore options they might not have considered.
That is the problem SAATHI AI is designed to explore.
✨ What I Built
- AI Safety Check
This is the central feature of SAATHI AI.
A user can describe a concerning situation and receive an AI-generated analysis of potential warning signs, along with practical suggestions.
For example, imagine someone says:
A person repeatedly asks for my location, gets angry when I spend time with friends, and expects me to reply immediately.
The AI may identify possible patterns involving digital monitoring, isolation, or boundary violations.
In another scenario, someone repeatedly appears at places the user visits, contacts them through different accounts, and makes intimidating statements.
The AI can highlight possible concerns such as repeated unwanted contact, surveillance, and intimidation.
I tested the AI Safety Check using example scenarios like these.
The purpose is not to label another person as guilty or claim that a crime has occurred. It is to help users reflect on the behaviour described and consider possible next steps.
- Practical Safety Suggestions
Recognizing a warning sign is only one part of the problem.
SAATHI AI can also suggest practical options, such as reviewing account privacy settings, keeping records of unwanted contact when safe to do so, contacting a trusted person, or exploring professional support.
The suggestions are advisory. What is appropriate depends on the person's circumstances, and an AI-generated recommendation may not always be safe or suitable.
- Confidential Help Request Interface
The prototype includes a Help Request interface that demonstrates how someone could seek additional support.
The idea is to make asking for help feel more approachable and respectful.
However, the current interface does not guarantee that a request reaches a real support team. A secure submission system and a real support workflow remain areas for further development.
- Human Review Dashboard Demonstration
Sensitive situations deserve more than automated decisions.
SAATHI AI includes a demonstration of a Human Review Dashboard to explore how a human-centred review process might work alongside AI.
The intended approach is for AI to assist with organizing information while people remain responsible for sensitive decisions.
At present, the dashboard is a demonstration. It is not connected to a verified team of advocates.
- Safety Resources and Quick Safety Exit
The website also includes a safety resources section and a Quick Safety Exit feature.
These features are intended to make safety information easier to access and help users leave the current page quickly when needed.
A quick exit cannot guarantee that browser history or other activity traces are erased. The accuracy and regional coverage of safety resources also need further verification.
🛠️ How I Built It
I used Google AI Studio to help develop the SAATHI AI web application.
The project repository includes technologies such as:
React for the user interface
TypeScript for application development
Vite for development and build tooling
Node.js for server-side functionality
Google AI Studio as the development environment
The project explores the use of open-weight AI, including Google's Gemma model family, for safety-related analysis.
The exact model configuration and integration should be confirmed from the source code. The important part of this project is exploring how AI-generated analysis can be presented in a way that encourages awareness without pretending to provide certainty.
I tested the AI Safety Check with example situations involving controlling behaviour and stalking. These tests helped me see how the system could identify possible warning signs and present suggestions.
The other components are at prototype or demonstration stage and require additional implementation and testing.
🌍 Why Does Open Innovation Matter?
Women's safety is not a problem that can be solved by one application or one developer.
It requires thoughtful design, careful testing, reliable information, and an understanding of the different circumstances people may face.
Open innovation creates opportunities for developers, researchers, designers, and communities to contribute different perspectives to a shared problem.
For a project like SAATHI AI, this matters in several ways.
First, transparency matters. Developers can inspect the code, understand implementation choices, identify weaknesses, and suggest improvements.
Second, experimentation becomes more accessible. Open-weight AI models give developers opportunities to explore how AI can be adapted for specific applications, subject to the model's license and technical requirements.
Third, collaboration can improve the project. Contributors could help with Marathi and Hindi support, accessibility, AI evaluation, privacy engineering, and the verification of safety resources.
Finally, responsible innovation matters as much as technical innovation. A safety-related application should not simply produce convincing answers. It should communicate uncertainty, respect personal choice, and avoid promises it cannot keep.
Open innovation does not automatically make a system safe or accurate. It creates opportunities to examine, challenge, and improve the work.
That is the kind of development community I hope SAATHI AI can benefit from.
🔮 What's Next?
SAATHI AI is a prototype, and there is still important work to do.
My future plans include:
Adding Marathi and Hindi language support.
Improving AI evaluation across different scenarios.
Developing a secure help-request workflow.
Exploring a real human-review process with appropriate safeguards.
Verifying emergency contacts and regional support resources.
Improving accessibility across devices.
Strengthening privacy and security.
Testing the application more thoroughly before any real-world use.
I especially want to explore how a safety-focused AI tool can remain useful without collecting unnecessary personal information or creating a false sense of security.
💭 What I Learned
Building SAATHI AI made me think beyond the question of whether AI can generate an answer.
In a sensitive application, the more important questions are:
Is the answer understandable?
Could it create a false sense of certainty?
Does it respect the user's choices?
Are the limitations clear?
What happens when the system gets something wrong?
These questions should influence the design from the beginning, not be added as an afterthought.
SAATHI AI is an early step in exploring these challenges. It is not a finished safety service, but it represents an idea I want to keep developing.
🔗 Explore SAATHI AI
Live Demo: https://saathi-ai-women-s-safety-platform.ai.studio
GitHub Repository: https://github.com/shraddhakolate30-maker/saathi-ai
I welcome constructive feedback, especially on responsible AI design, accessibility, privacy, and ways to make the project more useful.
If you have suggestions for improving SAATHI AI, I would love to hear them.
SAATHI AI is an educational prototype. AI-generated analysis may be inaccurate and does not replace emergency services, trained professionals, or individualized support. The project does not determine guilt or guarantee safety. Do not enter identifying or highly sensitive information into an unverified demo.
Top comments (0)