P
PENSIVE Sovereign Action Rails
SOVEREIGN ACTION INFRASTRUCTURE

PENSIVE

Building the Sovereign Autonomous Action Layer for India's Digital Infrastructure

Co-Founders: Mano Sanjay & Keshav

Pensive replaces bloated 70B+ Western LLMs with domain-native 3B Small Language Models paired with deterministic protocol guards, enabling autonomous agents to execute on ONDC, OCEN, and Account Aggregator rails at just ~8 paise per transaction.

Launch Protocol Simulator Calculate ROI (₹ Lakhs & Cr)
98.4%
Protocol Accuracy
vs 42.1% Frontier LLMs
8 Paise
Cost per Action
vs ₹2.60 Frontier Models
<200ms
Execution Latency
Real-time edge response
50,000
Sandbox Traces
ONDC, OCEN, & AA rails

The DPI Execution Deficit

Why traditional Western frontier models break down when exposed to India's population-scale asynchronous protocols.

Western LLMs Fail on DPI

Global frontier models excel at text processing but fail on India's asynchronous Digital Public Infrastructure (ONDC, OCEN, AA). Multi-party webhook states cause severe syntax errors, dropped transaction contexts, and schema hallucinations.

Beckn Protocol Validation Error

RFC 3339 timestamp dropped; 'bpp_id' hallucinated; asynchronous /on_search callback timed out after 30,000ms.

Unviable Micro-Economics

Querying 70B+ general-purpose models for standard protocol execution costs ~₹2.60 ($0.03) per API transaction. For high-velocity Indian commerce and payments, this unit economic barrier prevents enterprise adoption.

10 Lakh (10,00,000) Monthly Transactions:

Frontier LLMs: ₹25.95 Lakh/mo vs Pensive: ₹86,500/mo.

ARCHITECTURAL BREAKTHROUGH

Pensive Dual-Core Engine

Combining domain-native Small Language Models with deterministic non-neural validation to guarantee 100% protocol adherence.

Intent Parser SLM

Fine-tuned 3B–8B Small Language Model explicitly tokenized to parse natural language intent into structured API schemas without hallucinating extraneous parameters.

Domain-Specific Tokenizer

Deterministic Guard

Non-neural validation layer enforcing OpenAPI and Beckn protocol rules before any payload touches the live network. Rejects invalid schemas in <5ms.

100% Schema Conformance

Async Resilience

State-machine tracking that manages multi-turn webhook callbacks, network timeouts, and consent gates (Aadhaar / Account Aggregator handles) with guaranteed delivery.

Zero Dropped Webhooks

Execution Pipeline Flow

1
Natural Intent
Agent / User Prompt
"Order 2kg rice"
2
Intent Parser SLM
3B Domain-Native Model
Generates Beckn Schema
3
Deterministic Guard
OpenAPI / Beckn Rules
0% Schema Errors
4
National Rails
ONDC / OCEN / AA
<200ms Execution
POPULATION-SCALE ECOSYSTEM

Unlocking the DPI Economy

Pensive provides the autonomous intelligence layer connecting enterprise software directly into national rails.

From ONDC product discovery to instant OCEN credit origination and Account Aggregator data pipelines, we convert complex multi-party protocols into programmable single-step AI actions.

ONDC (Open Network for Digital Commerce)

Retail & Logistics

Autonomous item discovery, buyer-app order initiation, multi-party checkout, and logistics assignment executed through Beckn protocol callbacks.

OCEN (Open Credit Enablement Network)

Credit Origination

Automated MSME loan applications, underwriting checks, lender handshakes, and instant loan disbursement verification in a single AI invocation.

Account Aggregator (AA) Ecosystem

Consented Data

Consented financial data pipelines, automated OTP handling, bank statement reconciliation, and financial information user (FIU) telemetry.

RADICAL UNIT ECONOMICS

Radical Efficiency at Scale

By replacing bloated 70B+ parameter models with our domain-native 3B Action SLM, Pensive slashes execution costs to ~8 Paise (₹0.08) per transaction while keeping latency under 200 milliseconds.

Interactive Model

DPI Micro-Economics Calculator

96.7% Cost Slashed
Monthly Transaction Volume: 10 Lakh tx/mo
10K 1 Lakh 10 Lakh 50 Lakh 1 Cr 2.5 Cr+
Frontier 70B+ LLM ~₹2.60 / tx
₹25.95 Lakh
Average Latency: ~1,850ms
Pensive Dual-Core SLM ~8 Paise / tx
₹86,500
Average Latency: <180ms
Estimated Monthly Savings
₹25.08 Lakh/mo ($29,000/mo)
Annual Net Impact
₹3.01 Cr ($348,000)
LIVE PROTOCOL PLAYGROUND

Experience the Engine in Action

Select an intent or test protocol generation across ONDC, OCEN, and Account Aggregator.

1. Intent Parser SLM
Schema tokenization
2. Deterministic Guard
OpenAPI validation
3. Async Resilience
Webhook handshake
pensive-engine --live-trace
Payload Verified Latency: 148ms Cost: ₹0.08 (8 Paise) 

          


          
          

            
            Frontier Model Deficit: Multi-party webhook state dropped and schema hallucination observed.
          


        


      


    


    
    

      
      

        

          RIGOROUS BENCHMARK EVALUATION
        

        

          Protocol Execution Accuracy
        

        

          Benchmark tested across 50,000 synthetic & live sandbox execution traces on ONDC, OCEN, and Account Aggregator specifications.
        

      


      
      

        
        

          
          
          

            

              
                
                Pensive Action SLM
              
              98.4%
            

            

              

            

            
Zero schema hallucinations; 100% Beckn transaction ID binding pass rate.

          


          
          

            

              Frontier General LLM (70B+)
              42.1%
            

            

              

            

            
Fails on multi-party async callbacks; hallucinated parameters in 48% of traces.

          


          
          

            

              Generic Indic Chatbot
              28.6%
            

            

              

            

            
Unstructured responses; lacking deterministic protocol guard integration.

          


        


        
        

          

            
              

                Failure Mode
                Frontier Models
                Pensive Dual-Core
              

            
            
              

                Beckn Schema Hallucination
                38.4% of invocations
                <0.01% (Guarded)
              

              

                Webhook Callback Dropped State
                49.2% on multi-turn
                0.00% (State-Machine)
              

              

                Consent Flow Rejections (AA)
                31.7% schema mismatches
                1.2% (Valid format)
              

              

                Median Execution Latency
                1,820ms
                164ms
              

            
          

        


      


    


    
    

      
      

        

          STRATEGIC TRAJECTORY
        

        

          Execution Roadmap
        

        

          From core RLVR training on staging sandboxes to sovereign enterprise stacks for Tier-1 institutions.
        

      


      
      

        
        
        

          

            Q1
          

          
Core RLVR

          

            Complete sandbox execution training on Setu, ONDC, and Beckn staging environments with 50,000+ synthetic traces.
          

          Active Milestone
        


        
        

          

            Q2
          

          
Design Pilots

          

            Deploy enterprise pilots with 10 NBFCs and ONDC logistics buyer applications in live closed-alpha environments.
          

          Upcoming
        


        
        

          

            Q3
          

          
Enterprise Stack

          

            Launch on-premise sovereign deployment for Tier-1 Indian banks and healthcare networks with strict data sovereignty.
          

          Upcoming
        


        
        

          

            Q4
          

          
API Platform

          

            Open usage-based developer platform for autonomous DPI agent orchestration with sub-millisecond gateway caching.
          

          Upcoming
        


      


    


    
    

      
      

        
        

        


        

          
          INVESTMENT OPPORTUNITY
        


        

          Join Us in Building Pensive
        


        

          The Sovereign Action Foundation for India's AI Future
        


        
        

          
Seed Round

          
Raising ₹12.5 Crore

          
($1.5M Seed Investment)

          
          

            
Use of Funds Allocation

            
            

              

              

              

            


            

              

                
45% (₹5.63 Cr)

                
GPU Compute & RLVR

              

              

                
35% (₹4.37 Cr)

                
Core Engineering

              

              

                
20% (₹2.50 Cr)

                
GTM Pilots

              

            

          

        


        
        

          
          Mano Sanjay & Keshav (Founders)
        


        
        

          
          
        


      


    


  


  
  

  
  

  
  

    

      
      Opening email client...