Research & Systems Architecture Lab
Ground-level engineering research solving fundamental computing bottlenecks across education, precision agriculture, edge artificial intelligence, and clean energy infrastructure.
Edge-First Local NPU vs. Central Cloud Roundtrip Latency
Sub-150ms facial matching & local SQLite persistence without traversing external web backbones. Zero packet drops.
Subject to cellular congestion, ISP downtime, TLS negotiation overhead, and third-party data center outages.
Active Research Focus Areas
Zero-Cloud Raft Consensus & Multi-PC Local Sync
Air-Gapped P2P Distributed Systems for Institutional Operations
Traditional cloud ERP architectures introduce severe latency and single-point-of-failure vulnerabilities when campus internet links fluctuate. Our research models a deterministic peer-to-peer Raft state machine running over standard 1Gbps Ethernet LAN, achieving zero-drift SQLite ledger replication with sub-5ms commit windows.
- Dynamic leader election over local mDNS discovery
- Automatic journal log compaction and delta snapshot compression
- Cryptographically signed transaction vectors for anti-tamper ledgers
Sub-150ms Biometric Ingress & Vector Hash Hiding
Edge Neural Processing for High-Throughput Turnstiles
Addressing morning ingress choke points at large campuses: by running quantized lightweight MobileFaceNet embeddings directly on local NPU acceleration, facial vectors are matched against 2,000 enrolled subjects in under 150 milliseconds. Raw photographic pixels are immediately discarded after vectorization, ensuring absolute statutory DPDP 2023 compliance.
- Edge-quantized 512-dimension mathematical embeddings
- Zero raw image persistence across persistent disk storage
- Hardware-level anti-passback and liveness depth verification
Deterministic EV Charging & Open Telemetry Feeder Protocol
Edge-First Charging Infrastructure & Auto-Unlock Sharing
Investigating national EV adoption bottlenecks: we design an autonomous edge state machine protocol that combines FASTag RFID pre-authorization with predictive 10-minute queue broadcast APIs and mechanical auto-unlock charging guns, eliminating server overload and charger hogging across highway corridors.
- Offline-first OCPP local transaction buffer with 2G/Satellite sync
- Automated host profit-share splitting via national payment rails
- Underground disaster-resilient high-voltage commercial feeder design
Crop Care: Early Disease Spectral Isolation & Yield Optimization
Multispectral Plant Pathology & Low-Bandwidth Agronomic AI
Combining sub-surface LoRaWAN soil moisture probes with mobile leaf pathology models to diagnose crop rusts, borers, and blights before macroscopic wilting occurs. Designed to function completely offline on low-cost smartphones with regional Hindi audio instructions.
- Real-time visual inference for Sugarcane Red Rot & Wheat Yellow Rust
- Automated NPK & bio-fertilizer dosage calculator calibrated per acre
- Voice synthesis engine delivering instant agronomic advice in regional languages
Cosmic Convergence & Sovereign Deep Space Machines
Non-Volatile Compute Architectures & Post-Planetary AI
A fundamental inquiry into the endurance of artificial intelligence beyond terrestrial boundaries. We investigate radiation-tolerant state machines, self-governing robotic resource convergence, and the philosophical imperative of preserving human civilization data in non-volatile deep space nodes.
- Autonomous consensus over variable light-speed communication delays
- Self-compiling micro-kernels for long-duration deep space probes
- Deterministic ethics and non-coercive governance algorithms