Badrinarayanan
Rangarajan
ORG MERCEDES-BENZ R&D INDIA
LOC TAMIL NADU · IN
EXP 18 YRS · 2 VENTURES · 4 PATENTS · 7+ PAPERS
I build AI systems under constraints — power, latency, certification, legacy. First it was flight computers on nano-UAVs; now multi-agent platforms across enterprise GPU clusters. Same discipline either way: the system has to work when the world stops matching the model.
Agents inside a 140-year-old enterprise.
The model is the easy part.
SEC 01 / SHEET 01
As Technology Program Manager for Enterprise AI at Mercedes-Benz R&D India, I architect multi-agent systems that operate inside legacy infrastructure that cannot be rewritten — 60+ engineering workflows, brownfield-first. When an agent fails inside a 30-year-old stack, it reformulates strategy from the failure trace instead of escalating to a human. We call this meta-cognitive recovery. Beyond the cloud: onboard ML for in-cabin sensing — INT8/FP16 on automotive gateways, ISO 26262 and GDPR compliant.
Trajectory: engineer → founder → architect.
SEC 02 / SHEET 01Non-linear by design. Banking systems taught scale. Drones taught physics and real-time limits. Founding taught the economics of deep tech. Mercedes is teaching what it takes to land AI inside an enterprise that predates the transistor.
Consultant · Tata Consultancy Services
- Technical lead and application manager for payment systems and fraud detection at major US financial institutions.
- Proved: software that handles real money has no tolerance for "mostly works."
Founding Member · UooLabs
- Designed and prototyped device-management components on OSGi & OMA-DM.
- Proved: zero-to-one is a different sport. Set the trajectory.
Researcher & Product Manager · NTU
- Built a Pixhawk-based flight computing platform for signal identification over unknown terrain.
- Published the meta-cognitive classifier lineage — PBL-McRBFN, McCIT2FIS — that runs through five peer-reviewed papers.
- Won research funding from A*Star, MINDEF, MOE.
Co-founder, Chief Architect & CTO · SwarmX
- Designed the full stack: ARM flight computers, precision landing, deep-learning analytics, cloud fleet management.
- Owned marketing, commercialization, and the architectural roadmap end to end.
Researcher · Republic Polytechnic
- Built a low-cost ARM drone navigation system as a lidar replacement for warehousing; tested at a Toyota facility.
- Led customer engagement and commercialization.
Technopreneur-in-Residence · ARTPARK & IISc
- Built OmniPilot — a modular AI flight computing platform with a cognitive decision-making layer for GPS-denied autonomy: navigation, failure recovery, real-time strategy adaptation. An agentic system before the word existed.
- Shipped nano-UAVs with custom AI flight computers and an acoustic counter-UAV defense stack.
Technology Program Manager, Enterprise AI · Mercedes-Benz R&D India
- 35–40% faster dev cycles; 60+ engineers shipping AI monthly; meta-cognitive recovery patterns for brownfield agents.
- Authoring the AURA agentic SDLC platform and an autonomous production line for software delivery — full dossiers on Sheet 02 →
- Responsible AI framework: SHAP/LIME explainability, audit trails, bias mitigation.
Nine cells. Each one shipped.
SEC 03 / SHEET 01To a paying customer, a regulated environment, or a peer-reviewed venue — no cell on this matrix is aspirational. The leverage lives in the seam: most enterprise AI leaders have never written firmware; most embedded engineers have never sized a GPU cluster.
Agentic Architecture
Multi-agent orchestration, LangGraph / AutoGen / CrewAI, meta-cognitive recovery, brownfield deployment.
Cognitive Systems
Meta-cognitive RBFN, McCIT2FIS, PBL-McRBFN — published research on cognitive architectures.
Distributed Inference
vLLM, Ray, Triton, DeepSpeed — GPU fleets orchestrated across nodes.
Edge AI & Embedded
ARM SOC/SOM, INT8/FP16 quantization, automotive gateways, ISO 26262.
Autonomous Systems
UAV flight computing, GPS-denied navigation, precision landing, swarm coordination.
Vision & VLMs
Real-time interior sensing, object detection, vision-language models for ADAS contexts.
Cloud & MLOps
Multi-cloud AI infrastructure, CI/CD for ML, DeepEval benchmarking, MLOps automation.
Responsible AI
SHAP / LIME explainability, audit trails, bias mitigation, GDPR-compliant deployment.
Build & Sell
P&L ownership, VC raise, enterprise & government sales, cross-cultural team building.
The drawing set.
SEC 04 / SHEET 01Five more sheets. Platforms with measured outcomes, an active research thesis, the full technical stack, an inference playbook, and a CFO-grade value model.
Three platforms, one thesis
AURA, the Customer Portal agents, GenUI — lean-canvas dossiers with schematics and measured outcomes. Plus three open-source repos and the field archive: 15 tapes from the UAV years.
OPEN SHEET →Cognitive Integration Intelligence
Agents that observe behavior, accumulate memory, and treat failure as data. Rust persistent memory layer, edge benchmarks, five papers, four patents.
OPEN SHEET →The GenAI stack, end to end
Pretraining a 1.5B model from scratch, 256× H100 distributed training, vLLM serving, code RAG over 8.4M LOC, SLM fine-tuning, in-vehicle edge deployment.
OPEN SHEET →Enterprise inference, end to end
The four-rung serving ladder, the substrate techniques, and a side-by-side Azure / AWS / GCP playbook — down to the AUTOSAR V-model for AI in vehicles.
OPEN SHEET →~€520M a year, built bottom-up
A CFO-grade model of AI value across the vehicle development lifecycle — cost pool × evidenced uplift, net of the verification tax. Three sub-cases, each clearing a 6× return.
OPEN SHEET →Education
SEC 05 / SHEET 01Executive Business Administration
Indian Institute of Management, Bangalore
B.E., Computer Science & Engineering
Visvesvaraya Technological University