Anya — AI-Powered Nutrition Intelligence for India
Submission for API Innovate 2026 — AI FOR BHARAT 2026 This project complies with all hackathon rules: ASI-1 API integration, completed registration form, original code with proper licensing, GitHub repository, documented codebase, and README with setup instructions.
Inspiration
India produces enough food to feed its entire population, yet 35.5% of children under 5 are stunted and 57% of women are anemic (NFHS-5, 2019-21). This is not a food crisis — it's a nutrition intelligence failure, sitting squarely at the heart of UN Sustainable Development Goal 2: Zero Hunger.
"End hunger, achieve food security and improved nutrition, and promote sustainable agriculture."
When we analyzed NFHS-5 data across India's 640+ districts, we found a pattern demanding action:
| District | State | Child Anemia Rate | Distance from Nearest Metro |
|---|---|---|---|
| Alirajpur | Madhya Pradesh | 79.0% | 800 km from Mumbai |
| Nandurbar | Maharashtra | 74.8% | 400 km from Mumbai |
| Dahod | Gujarat | 72.1% | 350 km from Ahmedabad |
| Kiphire | Nagaland | 68.3% | 1,200 km from Kolkata |
These crises are predictable and preventable. The data exists — health surveys, food prices, weather patterns, PDS records — but it sits in silos, never connected, never predictive, never acted upon in time.
India spends over ₹1.5 lakh crore annually on food and nutrition programs (PDS, ICDS, Mid-Day Meals), yet outcomes remain poor because interventions are generic, reactive, and culturally disconnected.
What if one platform could predict which deficiencies will spike where, generate precision fortification strategies, and create culturally appropriate meal plans families will actually eat — all within their budget?
That became Anya, meaning "grace" in Sanskrit. Because every child deserves adequate nutrition, regardless of which district they were born in.
What It Does
Anya is an end-to-end AI-powered nutrition intelligence platform built specifically for India. It directly addresses three core targets within UN SDG 2:
| SDG 2 Target | Anya's Response |
|---|---|
| End hunger and ensure access to safe, nutritious food | Predicts malnutrition hotspots and delivers culturally appropriate meal plans |
| End all forms of malnutrition | Generates micronutrient-specific fortification strategies per district |
| Double agricultural productivity of small-scale food producers | Prioritizes indigenous and locally grown crops in all meal plans |
The platform operates as a three-stage pipeline:
| Stage | Purpose | Key Output |
|---|---|---|
| PREDICT | Identify who is at risk | Per-district, per-micronutrient risk scores |
| FORTIFY | Determine what nutrients are needed | Cost-optimized fortification strategies |
| NOURISH | Deliver meals people will actually eat | 7-day culturally aware meal plans |
Stage 1: PREDICT
The prediction engine fuses four government data sources into a unified risk model:
| Data Source | What It Contributes |
|---|---|
| NFHS-5 | District-level stunting, wasting, and anemia prevalence |
| Agmarknet | Real-time food prices at mandi (market) level |
| IMD | Seasonal and weather data affecting crop availability |
| PDS Records | Public distribution system coverage and entitlement data |
Outputs:
- Per-district risk scores for five critical micronutrients: Iron, Vitamin A, Zinc, B12, and Vitamin D
- Emerging hotspot identification, flagging districts before crises escalate
- Coverage of India's top 10 most malnourished aspirational districts, representing 1.6 crore citizens
Stage 2: FORTIFY
The fortification engine generates cost-optimized intervention strategies per district using linear programming:
| Fortification Vehicle | Target Deficiencies | Delivery Mechanism |
|---|---|---|
| PDS Rice | Iron, B12 | State-run ration shops |
| Edible Oil | Vitamin A, Vitamin D | Universal cooking medium |
| Double-Fortified Salt | Iron, Iodine | Daily household consumption |
| Mid-Day Meal Flour | Zinc, Iron | School-based feeding programs |
Key features:
- Maximizes deficiency reduction per rupee spent, subject to FSSAI safety limits
- Incorporates bioavailability interaction constraints (iron is not paired with calcium inhibitors within the same meal window)
- Projects 3-month outcomes with estimated percentage reduction in anemia, Vitamin A deficiency, and zinc deficiency
Stage 3: NOURISH
This is where Anya diverges from every existing nutrition platform. The meal generator produces 7-day culturally aware meal plans accounting for:
| Parameter | Details |
|---|---|
| Cultural Group | 9 groups: Tribal Hindu, Upper-Caste Hindu, OBC Hindu, Muslim, Christian, Jain, Sikh, Buddhist, Mixed/Secular |
| Season | Adjusted for monsoon, winter, and summer crop availability |
| Budget | Plans generated for budgets as low as ₹2,500/month |
| Family Composition | Pregnant women, lactating mothers, infants (6–24 months), children, elderly |
| Local Availability | Only ingredients present in the district's markets and PDS entitlements |
The system prioritizes indigenous superfoods that outperform expensive supplements:
| Indigenous Food | Key Nutrients | Region |
|---|---|---|
| Ragi (Finger Millet) | Calcium, Iron | South and Central India |
| Moringa (Drumstick Leaves) | Vitamin A, Iron, Calcium | Pan-India |
| Wild Amaranth | Iron, Protein, Folate | Tribal regions |
| Red Ant Chutney | Iron, Protein | Chhattisgarh, Jharkhand |
| Jackfruit Seeds | Zinc, Protein | South and Northeast India |
Our model calculates it costs just ₹847 per person to shift them out of nutritional risk through this pipeline, compared to ₹3,000+ for clinical supplementation programs.
How We Built It
Architecture
| Layer | Technology | Role |
|---|---|---|
| Frontend | React.js | Command center dashboard with dark theme |
| Mapping | Mapbox | India district-level heatmaps for risk visualization |
| Visualization | D3.js | Interactive charts for nutrient breakdowns and trend analysis |
| Backend API | Python, FastAPI | Serves prediction, fortification, and meal generation engines |
| AI Integration | ASI-1 API | Powers core intelligence pipeline — risk reasoning, fortification logic, and meal plan generation |
| Risk Model | XGBoost | Trained on real NFHS-5 district data fused with price indices and poverty indicators |
| Optimizer | PuLP, SciPy | Linear programming engine maximizing nutrition outcomes per rupee |
| Meal Engine | Custom constraint satisfaction engine | Combines ICMR food composition data, cultural rules, market pricing, and PDS data |
| Database | PostgreSQL | Stores district profiles, food databases, and generated plans |
Data Sources
| Source | Granularity | Variables Used |
|---|---|---|
| NFHS-5 | District (640+) | Stunting, wasting, anemia, dietary diversity, maternal health |
| Agmarknet | Mandi/Market | Daily prices for 300+ food items |
| IMD | District | Rainfall, temperature, drought indicators |
| FSSAI | National | Fortification standards and safety guidelines |
| ICMR | National | Food composition tables (600+ foods), RDA values by age/gender |
| Soil Health Card Portal | Block | Soil nutrient data influencing regional crop profiles |
All data is sourced from real government open data portals. No synthetic or mock datasets were used.
Challenges We Ran Into
1. Modeling the Complexity of Indian Food
A single dish like dal varies in ingredients, nutritional content, and cost across every state — sometimes every district. Building the cultural dietary rules database for 9 groups required extensive research into permitted/prohibited ingredients, pregnancy-specific taboos, fasting calendars, and regional variations.
2. Reconciling Geographic Granularity
NFHS-5 reports at district level (640+), while Agmarknet prices are at mandi level (7,000+ markets). Mapping mandi prices to districts while accounting for transport costs required a custom spatial reconciliation layer.
3. Bioavailability in Meal Optimization
Iron absorption varies by 3–10x depending on meal composition:
$$ \text{Absorbed Iron} = \text{Dietary Iron} \times f(\text{enhancers}, \text{inhibitors}) $$
We built a nutrient interaction matrix enforced as a hard constraint — iron-rich foods are never scheduled alongside calcium-heavy ingredients.
4. Multi-Constraint Real-Time Optimization
| Constraint | Requirement |
|---|---|
| Nutritional Adequacy | $\sum_{i} n_{ij} \geq \text{RDA}_j \quad \forall j \in \text{micronutrients}$ |
| Cultural Compliance | No prohibited ingredients for the specified cultural group |
| Local Availability | Only ingredients in the district's seasonal food basket |
| Budget Limit | $\sum_{i} c_i \cdot q_i \leq B_{\text{monthly}}$ |
| Bioavailability | No antagonistic nutrient pairings within the same meal |
5. Incomplete District Data
Several districts had partial NFHS-5 data. We built fallback estimation logic interpolating from neighboring districts weighted by demographic similarity, with flagged uncertainty in the UI.
What We Learned
Malnutrition in India is a systems failure, not a food shortage. The food, programs, and budget exist. What's missing is the intelligence layer connecting prediction to intervention to delivery.
Cultural sensitivity determines adoption. A plan that tells a Jain family to eat eggs will be ignored. Anya works because it respects what people already eat.
Indigenous foods are nutritional powerhouses. Ragi, moringa, wild amaranth — these traditional foods consistently outperform commercial supplements in cost-effectiveness and bioavailability.
Bioavailability is the invisible crisis. India's anemia rates stay high despite iron supplementation because absorption, not intake, is the bottleneck. Proper meal pairing increases iron absorption by up to 6x.
India's open government data is powerful. NFHS-5, Agmarknet, IMD, FSSAI — the raw materials for intelligent public health intervention are freely available. They just need to be unified.
What's Next for Anya
| Phase | Milestone | Description |
|---|---|---|
| 1 | Full India Coverage | Expand to all 640+ districts with real-time Agmarknet integration |
| 2 | POSHAN Tracker Integration | Connect with government Anganwadi monitoring to reach 1.4M frontline workers |
| 3 | Anganwadi Mobile App | Offline-first, voice-based interface in 12+ languages with camera-based MUAC screening |
| 4 | Nandurbar Pilot | Deploy across 50 Anganwadi centers serving ~2,000 children |
| 5 | WhatsApp Delivery | Personalized weekly meal plans via voice messages in local languages |
| 6 | Dynamic Pricing | Real-time meal plan adjustment when local ingredient prices spike |
| 7 | Open Source Release | Publish the platform for adaptation by other countries |
Social Impact
The Problem at Scale
| Indicator | Value | Source |
|---|---|---|
| Children under 5 stunted | 35.5% (4.87 crore) | NFHS-5 |
| Women 15–49 anemic | 57.0% (31.8 crore) | NFHS-5 |
| Children under 5 wasted | 19.3% (2.65 crore) | NFHS-5 |
| Annual GDP loss from malnutrition | 4% (~\$68 billion) | World Bank |
How Anya Creates Impact
- Predictive, not reactive — identifies at-risk districts before malnutrition spikes
- Affordable at scale — ₹847 per person vs. ₹3,000+ for clinical supplementation
- Culturally grounded — builds plans around what families already eat and believe
- Empowering frontline workers — puts actionable intelligence in the hands of 1.4M Anganwadi workers
Projected Impact (Nandurbar Pilot)
| Metric | Current | Projected (6 months) |
|---|---|---|
| Child anemia prevalence | 74.8% | 58–62% |
| Dietary diversity score | 3.2/9 | 5.5–6.0/9 |
| Monthly nutrition cost/family | ₹3,200 | ₹2,500 |
| Families with culturally appropriate plans | 0% | 100% of pilot |
SDG 2 Alignment
| Target | Anya's Contribution |
|---|---|
| 2.1 End hunger, ensure year-round access to nutritious food | Budget-optimized meal plans using local, seasonal ingredients |
| 2.2 End all forms of malnutrition | Micronutrient-specific prediction and fortification for stunting, wasting, anemia |
| 2.3 Double small-scale food producer productivity | Drives demand for indigenous and locally grown crops |
| 2.4 Sustainable food production | Prioritizes seasonal, regional ingredients over imported/processed foods |
Anya doesn't propose a new food program. It's the missing intelligence layer that makes India's existing programs — PDS, ICDS, Mid-Day Meal Scheme, POSHAN Abhiyaan — work as intended. The food is there. The workers are there. The data is there. Anya connects them.
Log in or sign up for Devpost to join the conversation.