ARTIFICIAL INTELLIGENCE

The Intelligence Layer

Zero-shot time series prediction using Amazon Chronos and our proprietary Spatial Parallel Search.

Spatial Parallel Search

With limited temporal data, we cannot rely solely on historical regression. Our insight: Space can substitute for Time.

If Region A (Target) matches Region B (Source) in November, then Region B's December outcome is a strong predictor for Region A's future.

Algorithm: find_spatial_parallels

Vectorization: Convert target PIN's history into a 3D Pulse Vector [Growth, Flow, Labor].
KNN Search: Scan 19,000+ PINs to find "Parallel Universes" using Cosine Similarity.
Trajectory Cloning: Project the future trajectory of the top-k parallels onto the target.

cos(θ)

Vector Similarity

src/attestation.py: Attestor.chat

def chat(self, message, context):
  # 1. Format Data Context
  data_block = self._format_context(context)
  
  # 2. Inject into System Prompt
  full_prompt = f"""
    ## Current Data Context
    {data_block}
    
    ## User Question
    {message}
  """
  
  # 3. LLM Generation
  return self.client.chat.completions.create(
      model="minimax-m2.1",
      messages=[{
        "role": "user", 
        "content": full_prompt
      }],
      temperature=0.4
  )

The Attestor Engine

The Attestor class (src/attestation.py) is the brain of the chatbot. It uses a specialized System Prompt to enforce a "Weather Forecast" persona.

Intent Parsing

Uses regex/LLM to extract structured intents from natural language.

"What about Delhi?" → { location: 'Delhi', intent: 'forecast' }

Context Injection

The ContextEngine injects seasonality drivers into the LLM prompt.

System: "It is Harvest Season in Punjab. Expect high labor movement."