SAARTHI.AI

SAARTHI.AI — Your Health, Alive.

Inspiration

Health data is everywhere — wearables, ECGs, health reports, lifestyle patterns and environmental signals — but most platforms stop at showing dashboards and historical trends.

We wanted to answer a more useful question:

"What happens if I change something today?"

That led us to reimagine SAARTHI.AI as a living Digital Twin of a person's health. Instead of simply telling users what happened, SAARTHI lets them explore possible futures, understand the impact of interventions and then compare those predictions with what actually happens.

The inspiration was to make health intelligence feel less like another dashboard and more like a conversation with a system that understands your past, models your future and evolves with you.

What it does

SAARTHI.AI creates a personalized Digital Twin from multimodal health data and uses Gemini to turn natural-language questions into actionable health simulations.

Users can:

Connect health and lifestyle data to build a continuously evolving health state.

Upload health reports, charts and other supported documents for multimodal analysis.

Ask questions naturally, such as "Why is my health trajectory getting worse?"

Ask "What if?" questions such as "What happens if I improve my sleep?"

Compare multiple intervention scenarios using SAARTHI's simulation engine.

Ask Gemini to identify the highest-impact changes under personal constraints.

Visualize current and simulated future states through an interactive Digital Twin.

Track predicted outcomes against observed outcomes.

Earn Health XP for meaningful progress and intervention adherence.

Interact conversationally with SAARTHI through voice where supported.

The core loop is:

SEE → UNDERSTAND → SIMULATE → ACT → OBSERVE → EVOLVE

How we built it

SAARTHI combines a deterministic health intelligence layer with Gemini's multimodal and agentic capabilities.

Gemini Intelligence Layer

Gemini 3.7 Flash acts as the reasoning and orchestration layer. It understands user intent, interprets multimodal context, selects the appropriate SAARTHI tools and converts natural-language requests into structured actions.

We use Function Calling to connect Gemini directly to SAARTHI capabilities such as:

Current health state

Historical trends

Intervention simulation

Trajectory comparison

Outcome evaluation

Structured Outputs ensure that Gemini's responses can safely drive Digital Twin states and frontend interactions instead of relying on uncontrolled free-form text.

SAARTHI Intelligence Core

The existing SAARTHI engines remain responsible for quantitative computation, including risk prediction, trajectory modelling and counterfactual simulation.

This creates a deliberate separation:

Gemini reasons and orchestrates. SAARTHI computes and validates.

Digital Twin

We use Three.js / React Three Fiber to render an interactive Digital Twin whose state is driven by actual SAARTHI outputs.

When a modeled trajectory improves, deteriorates or diverges from reality, the Twin changes accordingly.

Memory & Context

PostgreSQL, pgvector and RAG provide longitudinal context including health history, previous interventions and outcomes.

Google Cloud & Generative AI

Vertex AI provides the Google Cloud AI infrastructure for Gemini workflows.

Where useful, Nano Banana enables personalized visual storytelling, while Veo can turn selected simulated futures into short cinematic "Future Replay" experiences.

Challenges we ran into

Making Gemini genuinely useful

The biggest challenge was avoiding a superficial chatbot integration. Gemini needed to actually operate SAARTHI rather than simply generate explanations.

We solved this by exposing real SAARTHI capabilities through structured tools and using function calling to connect natural-language intent with actual simulations.

Keeping AI and numerical intelligence separate

We did not want an LLM generating health scores or inventing predictions.

The solution was to keep quantitative computation inside SAARTHI's existing models while using Gemini for interpretation, orchestration and interaction.

Making the Digital Twin meaningful

A 3D avatar can easily become visual decoration.

We therefore tied the Twin's state to actual health metrics, intervention results, trajectories and prediction-versus-reality outcomes.

Balancing complexity with simplicity

Behind the scenes, SAARTHI combines multiple AI and analytical components. The challenge was hiding that complexity from the user.

The final experience is intentionally simple:

Ask → Simulate → Decide → Act → See what happened.

Working within hackathon constraints

Integrating multimodal reasoning, tool calling, simulation, a Digital Twin and generative media within a limited development window required prioritizing the core interaction over unnecessary features.

Accomplishments that we're proud of

We turned SAARTHI into an interactive Digital Twin

Instead of another health dashboard, users can see their current state and explore modeled alternatives through a living visual representation.

Gemini actually operates the system

Gemini is not just generating text. It understands user intent, calls SAARTHI tools, orchestrates simulations and converts verified results into structured application state.

Natural language becomes an executable simulation

A question like:

"What if I improve my sleep for the next 30 days?"

can become an actual structured intervention, trigger the SAARTHI simulation engine and return a visual comparison.

Prediction meets reality

SAARTHI doesn't stop after making a prediction. It can compare the predicted trajectory with subsequent observations and use that feedback to update the Digital Twin.

We created a new interaction model for health

Instead of forcing users to interpret charts and navigate complex dashboards, SAARTHI lets them simply ask:

"What should I change?"

and explore the consequences before acting.

What we learned

AI is most powerful when it can act

A conversational model becomes significantly more useful when it can access real application tools and take structured actions rather than simply produce text.

Multimodality is valuable when it reduces friction

Reports, charts, wearable data and natural language shouldn't exist as isolated inputs. Gemini allows them to become part of a common contextual understanding.

Generative AI and deterministic systems have different jobs

LLMs are excellent at understanding, reasoning, orchestration and explanation. Numerical health models are better suited for deterministic calculations.

Combining both is more robust than forcing either technology to do everything.

A Digital Twin needs a feedback loop

A prediction becomes more meaningful when the system eventually asks:

"Was I right?"

That is why prediction-versus-reality is central to SAARTHI.

The best interface may not look like a dashboard

The most natural interaction with complex health data can simply be a conversation:

"Why?" "What if?" "Compare these." "Let's do it."

What's next for SAARTHI.AI

Future Replay with Veo

Allow users to turn a selected simulated trajectory into a short cinematic visualization of their modeled future.

Deeper Multimodal Intelligence

Expand the Digital Twin's understanding of health reports, wearable streams, lifestyle information and other supported signals.

Continuous Reality Feedback

Move from periodic assessments toward a continuously evolving loop where observed outcomes regularly test and refine predicted trajectories.

More Powerful Intervention Planning

Allow Gemini to compare larger combinations of interventions while considering personal constraints such as time, consistency and feasibility.

Longitudinal Digital Twins

Build richer long-term representations of how a person's health trajectory changes across months and years.

Responsible Health Intelligence

Continue strengthening grounding, validation, safety controls and clear separation between observed measurements, model outputs, simulated futures and AI-generated explanations.

SAARTHI.AI

Your Health, Alive.

Gemini understands. SAARTHI simulates. Your Digital Twin evolves.