The Majka Project Story

The Spark of Inspiration: The Personal Battle of the 4th Trimester

The postpartum journey is really tough. This time brings a lot of physical and mental changes. Every woman's recovery is different; there is no one-size-fits-all method. New moms are juggling late-night feedings, caring for the baby, meeting other responsibilities, dealing with intense physical healing, and facing serious mental health problems like postpartum depression. They need rest but also reliable information.

Current platforms do not meet these needs. They provide generic videos or complex content that require focus and free hands. Most importantly, they do not offer the understanding needed to address both physical recovery and mental health at the same time. We wanted to create Majka as a free, accessible companion that tells moms: you have enough on your plate; we’ll take care of the planning. Every mom deserves the best and safest support available.

What Majka Does

Majka is a smart companion that changes the chaotic postpartum experience into a clear, personalized healing journey. We tackle fragmented care by combining five essential functions:

Complete Data Privacy: Your entire personalized health profile is stored locally and privately, making sure your most sensitive data stays confidential.

Deep Evaluation: Evaluates your unique physical and mental state through a thorough, well-researched process.

AI Rehab Coach: Offers customized workout plans for different phases, created by the Gemini LLM, focusing on your specific recovery timeline.

Voice Over Guide: Gives hands-free audio instructions for your rehab workouts, making them easy to follow even when your hands are busy.

Safety First: Operates on a Critical Safety Guardrail that overrides all advice to provide life-saving instructions if severe symptoms are detected.

How we built it

We built Majka to eliminate fragmented care and data anxiety by integrating security and personalization into every layer.

Our complete tech stack was chosen for performance, security, and the ability to handle complex, asynchronous AI tasks:

Frontend- React 19 + Vite, CSS modules Backend & Chatbot API - FastAPI, Google Generative AI (Gemini) Data & Persistence - Supabase client (for secure health profile access) Guided Sessions - OpenCV + MediaPipe (for real-time pose estimation)

Challenges we ran into

We encountered several difficult, yet rewarding, challenges in engineering Majka to be the safest and most personalized coach available:

Safety and Robustness Guardrails: Our most critical challenge was prompt engineering to ensure robustness and setting up critical guardrails for safety. This required intense instruction tuning of the Gemini LLM and robust Python logic to ensure that if a red flag (like heavy bleeding, fever, or explicit distress) is detected, all other advice is instantly overridden by the single, life-saving instruction to contact a doctor.

Customizing Workouts (AI-Driven Process): Making the entire coaching process fully AI-driven required extensive logic to ensure the workouts were extensively customized as per each new mom's comfort level and healing phase.

Real-Time Calibration (MediaPipe Integration): Integrating MediaPipe into our pipeline proved challenging. We had to ensure this complex system could provide accurate, real-time pose estimation to validate exercise form, which is essential for home-based rehab.

Breathing Comfort Calibration: It was challenging to make the anxiety relieving breathwork assistant effective and safe. We had to solve how to accurately calibrate the breathwork according to each user's comfort and not just a generic timer.

Accomplishments that we're proud of

Our dedication to safety, customization, and accessibility resulted in several key achievements that define Majka:

The Critical Safety Guardrail : We successfully implemented a life-saving safety mechanism powered by the Gemini LLM that reliably detects high-risk symptoms (heavy bleeding, distress) and immediately overrides all advice with a clear call to action to contact a healthcare provider. This is the bedrock of trust in Majka.

True Personalized AI-Rehab : We achieved 100% AI-driven workout customization, moving beyond simple rule sets to intelligently synthesize complex intake data and the mother's comfort level into a unique, safe, and progressive rehabilitation plan.

MediaPipe Integration for Safety : We successfully integrated the complex MediaPipe library with our coaching pipeline, enabling real-time form correction during home workouts. This provides a crucial layer of physical safety and prevents potential injury, which is a significant technical feat.

Accessibility and Cost : We built a premium, highly complex AI system incorporating pose estimation and contextual intelligence while adhering to our commitment to keep the service completely free, ensuring every new mother can access the best care, regardless of financial barrier.

Holistic Support : We combined physical healing (AI Rehab) with mental wellness (Anxiety Relieving Breathwork) and informational clarity (Myth Busters) into a seamless platform, proving that comprehensive care is possible.

What we learned

Building Majka taught us critical lessons at the intersection of AI, health, and engineering:

The Supremacy of Safety Logic: We learned that for healthcare applications, the Python safety code must govern the LLM. Simply instructing Gemini was not enough; we had to build robust external logic (our Critical Safety Guardrail) to validate and override the AI's response before it ever reaches the user. This established that programmatic safeguards are superior to prompt-based instructions in life-critical scenarios.

Asynchronous is Essential for Context: We learned that retrieving vast amounts of personalized contextual data (intake, time since childbirth) and feeding it quickly to the LLM requires a high degree of asynchronous orchestration (FastAPI/httpx). High-performance computing is necessary to deliver true personalization without frustrating the user with slow responses.

Custom Calibration: We discovered that user comfort is not standardized. Successfully building the Breathwork Assistant taught us the nuanced engineering required to calibrate an automated process (breathing pace) based on highly subjective user-defined comfort levels.

What's next for Majka

Majka's next phase of development is focused entirely on maximizing accessibility and creating a seamless, supportive experience:

Touch-Free Voice Bot Integration: We plan to fully implement a hands-free, conversational voice bot powered by text-to-speech technology (like the optional ElevenLabs integration) and our existing Gemini context pipeline. This allows mothers to ask questions and receive instant, spoken advice without ever needing to look at or touch their phone.