Skip to content

Breed Knowledge Base & Farmer Advisory

Every prediction is enriched with practical, offline guidance for farmers and the general public — no retraining, no extra model, no network call.

breed prediction ──► src/breed_knowledge.py ──► advisory (gender × age)
                     data/breed_knowledge.json

Coverage: all 68 classifier breeds (50 cattle + 18 buffalo) plus the crossbred/alias classes holstein_friesian and jersey.

What each breed entry contains

Field Example
display_name "Gir"
local_names {"en": "Gir", "hi": "गीर", "ta": "கிர்"}
origin_region "Saurashtra — Junagadh, Amreli, Bhavnagar and Porbandar, Gujarat"
utility_type "Milch (Dairy)", "Draught (Work)", "Dual-Purpose", "Ornamental / Dwarf Dairy"
milk_yield daily litres, lactation kg, fat %, lactation length, β-casein note (numeric ranges included for income maths)
physical_markers 3 bullets for visual verification, derived from data/breed_traits.json
feeding_and_care green/dry fodder, concentrate, mineral mixture, water (kg or g/day)
climate_and_disease_resistance heat tolerance, tick/parasite resistance, disease notes, housing, smallholder suitability
estimated_market_price_inr "₹45,000 - ₹100,000" (numeric range included)
govt_schemes RGM, NLM, NADCP, KCC-AH, DIDF, ICAR conservation support
data_confidence "curated" or "regional_default"

Data policy. Curated entries follow NBAGR/ICAR breed descriptions and standard Indian feeding norms (2–2.5 % body-weight dry matter, 0.4 kg concentrate per litre of milk above 5 L). Rare breeds without reliable published production data use the utility template and are explicitly flagged regional_default — the UI tells the farmer so. Prices, yields and vaccination timings vary by farm and district, and every advisory ends with a disclaimer.

Advisory engine

get_farmer_advisory(species, breed_slug, gender, age_group) builds a context matrix. It is pure dictionary assembly (~0.1 ms), so the UI can re-render on a toggle without touching the image model.

Context Emphasis
Female + Adult Lactation management, expected yield/fat, concentrate top-up (0.4 kg/L above 5 L), calcium + mineral mixture, mastitis prevention, drying-off, calving interval, milk income estimate
Female + Young Heifer growth ration (400–600 g/day gain), target breeding weight, age at first calving by breed type, estrous/heat signs
Male + Adult Milk is hidden (not the relevant metric); draught work capacity, safe load, yoke/hoof care, breeding-bull and semen value
Male + Young Bull-calf vs draught-ox decision, castration timing, growth target, temperament handling
Calf (any gender) Colostrum in the first 1–2 h, 10 % body-weight milk, calf starter, deworming on day 10/30/60, FMD/HS/BQ calendar, brucellosis for female calves
Unspecified General breed briefing when the model's gender/age estimate is unavailable

Returned shape:

{
  "breed": {"species": "Cattle", "slug": "gir", "display_name": "Gir",
            "local_names": {"hi": "गीर", "ta": "கிர்"},
            "origin_region": "...", "utility_type": "Milch (Dairy)",
            "data_confidence": "curated"},
  "context": {"gender": "Female", "age_group": "Adult"},
  "headline": "Gir adult female — protect lactation ...",
  "priority_actions": ["...", "..."],
  "nutrition": {"green_fodder": "22 - 30 kg/day", "...": "..."},
  "health": {"vaccination": [], "watch_for": [], "housing": "..."},
  "production": {"milk": {...}, "draught": null},
  "breeding": {"heat_cycle": "...", "service_window": "..."},
  "economics": {"market_price": "₹45,000 - ₹100,000", "schemes": []},
  "physical_markers": ["...", "...", "..."],
  "schemes": ["..."],
  "warnings": ["..."],
  "disclaimer": "..."
}

REST API

GET /api/breed-knowledge/<species>/<breed>

curl http://localhost:5000/api/breed-knowledge/Cattle/Gir
curl http://localhost:5000/api/breed-knowledge/Buffalo/Nili-Ravi   # alias-tolerant

Returns {species, breed, knowledge, advisory, advisory_options, latency_ms}. 404 for an unknown breed/species pair, 200 otherwise.

POST /api/advisory-recalculate

curl -X POST http://localhost:5000/api/advisory-recalculate \
  -H "Content-Type: application/json" \
  -d '{"species":"Buffalo","breed":"murrah","gender":"Female","age_group":"Adult"}'

Returns {advisory, context, latency_ms}. Measured ~0.07 ms per call (200 calls averaged), well inside the < 10 ms UI budget. Both endpoints exist in the Flask server (server/routes/knowledge_api.py, registered in server/app.py) and in the standalone tester (test_model.py).

Attached to predictions

POST /api/predict now includes a knowledge block computed from the predicted breed and the estimated/confirmed gender and age:

{
  "top_breed": "Gir",
  "gender": {"short_label": "Female", "...": "..."},
  "age_group": {"short_label": "Adult", "...": "..."},
  "knowledge": {"headline": "Gir adult female — protect lactation ...", "...": "..."}
}

OOD inputs return {"skipped": true, "reason": "out_of_distribution"} instead of misleading advice. If the advisory ever fails it returns an error key and the prediction itself still succeeds.

Generated data file

data/breed_knowledge.json is created automatically on first use:

python -m src.breed_knowledge --regenerate    # rewrite all 68 + 2 crossbred entries
python -m src.breed_knowledge --stats         # coverage counts
python -m src.breed_knowledge --show gir
python -m src.breed_knowledge --show gir --advisory Female Adult

The loader only adds missing breeds, so hand-edited numbers survive a restart. --regenerate overwrites everything from the module.

Current coverage: 59 curated, 11 regional defaults.

Verifying

python scripts/verify_breed_knowledge.py     # all entries + all 8 contexts + latency
python scripts/verify_attributes.py          # prediction-level integration

Limitations

  • Figures are typical ranges for the breed, not this animal's measurement.
  • Market prices are indicative 2024–25 farm-gate ranges and move with locality, lactation stage, fat % and festival demand.
  • Vaccination schedules must follow the local State Animal Husbandry Department; NADCP timing can differ by district.
  • regional_default breeds (e.g. Bhelai, Ghumsari, Marathwada) have thin published data and are flagged as estimates.