Sakhi

COMMON SYMPTOMS , MOOD VS CYCLE PHASE, FLOW INTENSITY, PAIN TREND GRAPHS
WEIGHT TREND, SLEEP HOURS, PCOS PREDICTION,THYROID PREDICTION,DOCTOR SUMMARY IN PDF.
ASKING BASIC DATA ABOUT THE USERS AFTER SIGN UP
MOOD TRACKER, SLEEP GRAPH, MATERNAL RISK ASSESMENT, DOCTOR SUMMARY.
PDF SUMMARY OF THE OVERALL HEALTH CONDITIONS TO CONSULT A DOCTOR.
WEIGHT GAIN, FETAL MOVEMENT COUNT,APPOINMENTS, NUTRITION BREAKDOWN IN PREGNANCY MODE.
PREGNANCY PROGRESS, BLOOD PRESSURE, FASTING BLOOD SUGAR GRAPH ANALYSIS.
CHANGING THE MENSTRUAL MODE TO PREGNANT MODE.
SETTINGS ABOUT WHAT REMAINDERS NEEDED.
NOTIFICATION ABOUT REMINDERS WITH DEADLINES.
CHATTING WITH OUR AI ASSISTANT ABOUT MENSTRUAL HEALTH AND GETTING PERFECT GUIDANCE.
CHATTING WITH OUR AI ASSISTANT ABOUT PREGNANCY HEALTH AND GETTING PERFECT GUIDANCE.
INTRO PAGE - 2
WHAT IS INSIDE THE WEBSITE SAKHI
INTRO PAGE - 1

Inspiration

Women's reproductive health data—spanning menstrual cycle logs, PCOS markers, and thyroid conditions—is often complex, deeply interconnected, and traditionally siloed. We were inspired to build SAKHI (a holistic health-tech companion) to bridge the gap between tracking symptoms and understanding underlying patterns.

To ensure the experience feels deeply personalized, we introduced two distinct conversational profiles: Her Mode and Mom Mode. We recognized that a woman's health journey changes dramatically over time. The needs of someone tracking their cycle or managing PCOS are fundamentally different from someone navigating pregnancy. Her Mode is designed for general reproductive health, providing AI-driven insights on menstrual cycles, thyroid data, and PCOS risks. In contrast, Mom Mode adapts the platform to focus on prenatal care, fetal development, and maternal well-being. By separating these contexts, we ensure the AI's persona, knowledge base, and predictive models provide relevant, empathetic, and specialized support for every stage of a woman's life. Ultimately, we wanted to empower users with an intuitive conversational interface backed by rigorous AI analysis to provide real-time, actionable insights.

What it does

SAKHI is an AI-powered conversational agent that helps women track and manage their reproductive health. Through an intuitive chat interface, users can converse with Sakhi to receive personalized insights based on their health data. Sakhi intelligently identifies when a user is asking for general advice versus when they need data-driven predictions. By routing queries through a robust machine learning backend, SAKHI can predict cycle lengths, estimate PCOS risks, and offer guidance on thyroid management. The dual "Her Mode" and "Mom Mode" ensures the advice is always tailored to the user's specific life stage, offering empathetic and accurate responses ranging from standard menstrual health to week-by-week fetal development.

How we built it

We engineered a robust 3-tier architecture:

Frontend: A highly interactive Next.js App Router providing a seamless chat interface and dynamic user onboarding.
Backend & Orchestration: An Express middleware server that maintains conversation history and integrates with the Gemini 2.5 API using function calling to intelligently route queries.
Machine Learning Engine: A FastAPI microservice that hosts our trained predictive models.

When a user asks about their cycle or PCOS risk, Gemini detects the intent and triggers a tool call. Our Express server then passes the structured data to FastAPI. We utilized predictive modeling logically grounded in probability:

$$ P(Y=1|X) = \frac{1}{1 + e^{-(\beta_0 + \beta_1 X_1 + \dots + \beta_n X_n)}} $$

where $X_i$ represents key biomarkers such as weight, cycle regularity, and acne severity.

Challenges we ran into

Agentic Routing: Orchestrating Gemini to reliably emit tool calls instead of hallucinating natural language responses required precise system prompting and strict JSON schema definitions.
Data Standardization: Harmonizing diverse inputs from natural language chat into the strict numerical formats demanded by our .pkl models.
Latency & Integration: Connecting three distinct layers (Next.js, Node/Express, Python/FastAPI) and an external LLM API without introducing noticeable lag during a live chat experience.

Accomplishments that we're proud of

We are incredibly proud to have built a system that genuinely feels empathetic while remaining mathematically rigorous. Successfully implementing Gemini Function Calling to bridge natural language processing with our local FastAPI machine learning models was a major technical milestone. Furthermore, developing the dual-persona framework (Her Mode and Mom Mode) within a single unified chat interface allowed us to deliver a highly nuanced and personalized user experience.

What we learned

LLM Function Calling: We mastered the art of decoupling natural language processing from deterministic ML scoring by using Gemini as an intelligent router rather than a direct answer engine.
Microservice Design: We learned how to isolate complex Python data science dependencies away from our fast Next.js UI via RESTful APIs.
Empathic AI Design: Most importantly, we learned how to synthesize raw probabilistic outputs—like a $0.74$ probability of PCOS—into a compassionate, easy-to-understand message that encourages medical consultation without causing unnecessary panic.

What's next for Sakhi

Moving forward, we plan to expand SAKHI's capabilities by integrating with wearable health devices (like smartwatches and continuous glucose monitors) to provide even more granular, real-time insights. We also aim to expand Mom Mode with a more robust postpartum recovery module, and we hope to fine-tune our own localized LLM to ensure absolute data privacy for all our users' sensitive health information.