ankit.systems

Research Contributions

Low-level systems research in Trusted Execution Environments (TEEs), constraint-based resource optimization, and resource-bounded Edge AI models.

PhD Thesis Objective & Open Questions

My primary research focus is building trustworthy, zero-trust intelligent infrastructure. In volatile cyber-physical grids (e.g. autonomous vehicles, border defense nodes), edge nodes run critical machine learning models in host environments that cannot be verified.

“How do we verify execution integrity and data confidentiality on remote edge devices subject to physical tampering, host compromises, and tight latency/power boundaries?”

By co-designing silicon-enforced memory enclaves (Intel SGX, ARM TrustZone), mathematical operations solvers (CP-SAT), and knowledge-distilled dynamic neural networks, my goal is to deliver platforms that operate securely in hostile edge environments.

Research & Core Systems

Intel SGX Confidential Computing Backend

Hardware-Enforced Private Document Processing

System Architecture Topology

Untrusted ClientJava / JNI BackendHost RAM
ECALL / OCALL API→ TLB Flush →
Intel SGX EnclaveC++ AES-128-GCMEPC Enclave Page Cache
* Secure memory limit discovered at ~16MB EPC bounds via dynamic binary sweeps, paging up to 2.86 GB of encrypted data through it.DCsv2/DCsv3 VM Verified

Summary

A hardware-isolated processing backend utilizing Intel SGX enclaves to protect memory-intensive document analysis pipelines from privileged cloud threats.

Problem Context

Deep learning and document analysis workloads deployed in third-party clouds are vulnerable to data leakage from compromise of the hypervisor, OS, or cloud administrators.

Research Motivation

Provide absolute cryptographic guarantees of confidentiality and runtime integrity for sensitive document pipelines using silicon-level isolation.

System Architecture

Untrusted Java Backend -> JNI Interface -> Secure Intel SGX C++ Enclave. Enclave memory boundaries configured to respect a ~16 MB trusted memory (EPC) footprint.

Applied Methods

Designed low-level ECALL/OCALL interfaces. Implemented AES-128-GCM authenticated encryption inside the enclave. Used memory paging/swapping to fit far more data than the enclave's native EPC capacity without breaking isolation.

Observed Results

Successfully processed workloads up to 2.86 GB of encrypted data on Azure Confidential VMs (DCsv2/DCsv3) inside a ~16 MB trusted enclave, with zero security degradation.

Future Directives

Explore transition to ARM TrustZone and RISC-V Keystone for mobile edge security topologies.

Technology Stack

Intel SGXC++JavaJNIAzure Confidential VMsAWS KMSAES-GCM
Production Engineering

GeoVision Deterrence System

Smart India Hackathon National Winner — Government R&D Category

Research & Core Systems

5-Layer Universal Scheduling Engine

Operations Research at Loughborough University

Research & Core Systems

Tactical Edge Wildlife Analytics

Knowledge Distillation on Raspberry Pi & ESP32

Research & Core Systems

HawkEye UAV Platform

UAV Video Intelligence & Geospatial Detection Platform (Government R&D Program)

Research & Core Systems

SpectraSam

Hyperspectral Crop/Land-Cover Classification Desktop App