Skip to content

Latest commit

 

History

2 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

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. We don't have a food crisis. We have a nutrition intelligence failure.

When we analyzed NFHS-5 data across India's 640+ districts, we found something striking: in Alirajpur, Madhya Pradesh, 79% of children are anemic, just 800km from Mumbai. These crises are predictable and preventable, but India's nutrition response is always too late, too generic, and culturally disconnected.

India spends ₹28,000 crore annually on nutrition programs. But:

  • Nobody predicts where deficiencies will spike next. We discover malnutrition after children are already severely affected
  • Fortification is one-size-fits-all. Every district gets the same iron-fortified rice, whether their crisis is iron, Vitamin A, B12, or zinc
  • Meal recommendations are culturally tone-deaf. Telling a tribal family in Jharkhand to "eat a balanced diet" means nothing when they don't know what to buy, how to cook it, or how to afford it within ₹2,500/month

The data to solve this already exists: NFHS-5 health surveys, Agmarknet food prices, IMD weather data, FSSAI fortification standards, ICMR food composition tables. But it all lives in silos, never connected, never predictive, never actionable.

So we asked: 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 operates as a three-stage pipeline: Predict. Fortify. Nourish.

1. PREDICT: Who's at risk?

Our ML engine fuses data from NFHS-5, Agmarknet food prices, IMD weather, PDS records, and socioeconomic indicators to generate per-district, per-micronutrient risk scores for Iron, Vitamin A, Zinc, B12, and Vitamin D.

What makes this India-specific:

Factor Why It Matters How We Model It
Monsoon seasonality Hunger peaks pre-harvest (Jun-Sep) Seasonal decomposition in LSTM
Festival cycles Fasting periods (Navratri, Ramadan) affect nutrition Calendar feature encoding
Marriage season Food prices spike, families eat less Price elasticity modeling
Migration patterns Male migration leaves women/children vulnerable NREGA enrollment as proxy
Caste-based diet Vegetarian communities have different deficiency profiles Demographic stratification
Tribal vs non-tribal Tribal blocks have 2-3x worse outcomes Separate sub-models
PDS disruptions When ration shops close, malnutrition spikes PDS offtake anomaly detection

The system covers India's top 10 most malnourished aspirational districts with real data for 1.6 crore citizens, generating risk scores like:

  • 🔴 Iron: 79% CRITICAL (rice-dominant diet + low egg intake)
  • 🔴 Vitamin A: 65% CRITICAL (seasonal vegetable scarcity)
  • 🟡 Zinc: 44% MODERATE (stable)
  • 🟠 B12: 55% HIGH (predominantly vegetarian population)

2. FORTIFY: What nutrients are needed and how do we deliver them?

India's current approach is broken: fortify ALL rice with iron and distribute EVERYWHERE. A district with a Vitamin A crisis gets iron-fortified rice it doesn't urgently need. Tribal communities eating millets, not rice, get nothing. No cost optimization. Limited budget spread thin everywhere.

Anya generates precision fortification strategies per district using linear programming:

Vehicle Micronutrient Target Cost Projected Impact
PDS Rice (FRK) Iron + Folic Acid + B12 General population ₹0.73/kg ↓18% anemia
Anganwadi THR Zinc + Vitamin A Children 6-36 months ₹1.2/packet ↓31% VAD
Double Fortified Salt Iron + Iodine All households via PDS ₹2/kg ↓12% anemia
Edible Oil Vitamin A + D All households ₹0.50/liter ↓24% night blindness
Mid-Day Meal Flour Iron + Zinc + B-complex School children 6-14 ₹0.90/kg ↓15% anemia

The optimizer accounts for bioavailability science: iron pairs with Vitamin C for 2-3x absorption, but iron and calcium compete, so they get separated into different meals and delivery channels.

Projected outcomes for Alirajpur: Anemia 79% to 53% (↓33%), Vitamin A deficiency 65% to 42% (↓35%) at a cost of ₹12.4 lakhs/quarter.

3. NOURISH: Meals people will actually eat

India's food culture is the most complex in the world. 9 major religious/cultural dietary patterns. Regional staple variations. Pregnancy taboos that harm nutrition. Seasonal fasting calendars. Budget constraints as low as ₹2,500/month.

Anya generates 7-day culturally aware meal plans optimized across 8 simultaneous constraints:

  1. Nutritional: Meet WHO/ICMR RDA for each family member
  2. Cultural: Only include foods this family would actually eat (9 groups: Tribal Hindu, Muslim, Jain, Sikh, Hindu Brahmin, and more)
  3. Economic: Minimize cost while meeting nutrition targets
  4. Availability: Only use ingredients available at local market + PDS + seasonal wild foods
  5. Practical: Recipes cookable on a single-burner chulha, no fridge needed
  6. Fortification-aware: Account for nutrients already in fortified rice/salt/oil
  7. Bioavailability: Pair iron-rich foods with Vitamin C sources, separate calcium inhibitors
  8. Lifecycle: Different portions for pregnant mothers vs infants vs elderly

The system highlights indigenous superfoods that outperform expensive supplements: red ant chutney (iron-rich), ragi (calcium + iron), moringa (Vitamin A), wild amaranth, drumstick leaves, mahua flowers, bamboo shoots, and wild mushrooms (B12).

Our model calculates it costs just ₹847 per person to shift them out of nutritional risk through this pipeline.


How we built it

Component Technology Purpose
Frontend React.js, D3.js, Mapbox Dashboard with India district heatmaps and interactive visualizations
Backend Python, FastAPI RESTful API serving prediction, fortification, and meal engines
Risk Prediction XGBoost + LSTM (PyTorch) Multi-source data fusion for per-micronutrient risk scoring
Fortification Optimizer PuLP, SciPy Linear programming maximizing deficiency reduction per rupee spent
Meal Generator Constraint satisfaction + LLM Multi-constraint optimization with cultural rules database
Data Layer PostgreSQL, Redis Structured storage + fast caching
Data Sources NFHS-5, Agmarknet, IMD, FSSAI, ICMR Real government open data, no synthetic datasets

The ML architecture uses an ensemble approach: XGBoost handles static demographic and geographic features, while LSTM captures temporal patterns (seasonal hunger, food price trends, weather). A weighted fusion layer combines both into per-micronutrient risk scores with confidence intervals and causal attribution.

The fortification optimizer solves a linear programming problem: maximize nutrition impact per rupee, subject to FSSAI safe upper limits, bioavailability interaction constraints, existing fortification programs, supply chain feasibility, and population dietary patterns.

The meal generator runs a constraint satisfaction engine against the Indian Food Composition Tables (ICMR), a cultural rules database covering 9 religious/cultural groups, local market pricing from Agmarknet, PDS entitlement data, and WHO/ICMR RDA values. It produces weekly plans with per-meal costs, nutrient coverage percentages, bioavailability tips, and food group breakdowns.


Challenges we ran into

  • Indian food is incredibly complex to model. A single dish like "dal" varies in ingredients, nutrition, and cost across every state. Building the cultural rules database for 9 religious/cultural groups with accurate dietary restrictions, pregnancy taboos, fasting calendars, and indigenous food knowledge was the hardest non-technical challenge.

  • Bioavailability modeling. Iron absorption varies 3-10x depending on what else is in the meal. Building the micronutrient interaction matrix (iron + Vitamin C = enhances, iron + calcium = inhibits, zinc + iron = competes) and enforcing it across both fortification recommendations and meal generation was scientifically demanding.

  • Data inconsistency across government sources. NFHS-5 has district-level data but Agmarknet food prices are at mandi (market) level. Reconciling geographic granularity across health, agriculture, weather, and economic datasets required careful mapping and normalization.

  • Multi-constraint optimization in real-time. Generating a meal plan that simultaneously meets nutrition targets, respects cultural rules, uses only locally available foods, accounts for fortified food inputs, optimizes bioavailability pairings, differentiates by family member lifecycle stage, AND fits a ₹2,500/month budget is a hard optimization problem. Getting it to run in under 2 seconds required careful algorithmic design.

  • Addressing cultural sensitivity without being patronizing. Pregnancy myths ("don't eat papaya", "eat less so baby is small") actively harm nutrition. We had to design a myth-busting approach that respects elders and frames corrections as "new knowledge" rather than "you're wrong", because in Indian families, the mother-in-law is often the key dietary decision-maker.


What we learned

  • Malnutrition in India is not a food problem. It is a data, delivery, and cultural sensitivity problem. The food exists. The government programs exist. What's missing is the intelligence layer connecting prediction to action.

  • Cultural context is non-negotiable. A nutrition solution that ignores caste, religion, regional food habits, and pregnancy taboos will fail in India, no matter how technically impressive it is. We built support for 9 distinct cultural groups because one-size-fits-all kills adoption.

  • Indigenous foods are nutritional powerhouses. Red ant chutney (iron), ragi (calcium + iron), moringa (Vitamin A), wild mushrooms (B12) outperform expensive supplements but have been systematically overlooked by modern nutrition programs. The tribal diet is often already nutritious. The problem is consistency, not content.

  • Bioavailability matters as much as quantity. You can eat enough iron on paper but absorb almost none if paired with the wrong foods. This invisible factor is a major driver of India's anemia crisis and almost no existing tool accounts for it.

  • Government open data is a goldmine. NFHS-5, Agmarknet, Soil Health Cards, IMD, ICMR Food Composition Tables. The infrastructure for intelligent nutrition planning exists. It just needs to be connected.

  • Fortification without personalization is waste. Sending the same fortified rice everywhere means districts with Vitamin A crises get iron they don't urgently need, while their actual deficiency goes unaddressed. Precision matters.


What's next for Anya

  • Expand to all 640+ districts with full NFHS-5 coverage and real-time Agmarknet price feeds
  • Integrate with POSHAN Tracker, the government's existing Anganwadi monitoring platform, so Anya's predictions and meal plans reach 1.4 million Anganwadi workers directly
  • Anganwadi worker mobile app with offline-first architecture, voice-based UI in 12+ Indian languages, and camera-based MUAC measurement for instant child malnutrition screening
  • Pilot in Nandurbar, Maharashtra with district ICDS office across 50 Anganwadi centers serving ~2,000 children
  • WhatsApp-based meal plan delivery sending personalized weekly plans to mothers via voice messages in their local language
  • Real-time food price integration dynamically adjusting meal plans when local ingredient prices spike
  • Open-source the platform so other countries with similar challenges can adapt Anya to their context

Built With

react python fastapi xgboost pytorch mapbox d3js pulp scipy postgresql redis nfhs-5 agmarknet

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages