ankit.systems
Systems & Security Research Identity

Ankit Kumar

Undergraduate Systems Researcher & Software Engineer

Tailor my profile for your focus:

Applied ML & Computer Vision • Confidential Computing • Edge AI • Scalable Architecture

Ankit Kumar

Research & Systems Engineering

I build applied machine learning and computer vision systems, from detection and segmentation models to LLM-backed platforms, and engineer the secure, high-performance infrastructure to run them in production. My work spans confidential computing, edge AI, and distributed systems.

My work bridges the gap between academic research and industrial execution, spanning applied deep learning, hardware-rooted security, optimization models, and high-concurrency web platforms with experience across the United States, the United Kingdom, and India.

Read my full story & philosophy

Academic Trajectory & Systems Focus

My academic path is defined by a strong upward trajectory in advanced systems engineering and computer science coursework. In my recent semesters, I pushed my performance to score consecutive 8.79, 9.00, and 8.65 SGPAs, demonstrating a clear alignment between theoretical academic rigor and hands-on systems implementation.

Current Research & Engineering Problems

Active open problems, system constraints, and theoretical puzzles I am currently exploring:

TEE-Bound Edge Runtimes

How can we protect YOLO execution and motion matching Kalman loops from physical side-channel compromises on remote edge nodes without relying on power-hungry TPMs or active cloud links?

Distributed Offline Attestation

How do distributed ad-hoc mesh networks establish and verify computational integrity signatures when nodes must run entirely offline in denied cyber-physical zones?

NP-Hard Pruning Heuristics

Can we leverage reinforcement learning policies to automatically generate availability and distance pruning heuristics for integer solvers like CP-SAT, avoiding hand-crafted matrix structures?

Research Chapters

Explore full papers
🇺🇸NJIT

Intel SGX Enclaves

Secure Computing

Programmed ECALL/OCALL runtime isolation for document indexing. Measured Enclave Page Cache (EPC) limits on Azure DCsv2/DCsv3 VMs.

🇬🇧Loughborough

Optimization & OR

Decision Support

Engineered scheduling constraint solvers using Google OR-Tools CP-SAT. Applied spatial distance heuristics to output allocations in <2.3s.

🇮🇳Govt. R&D / SIH

Secure Geo-Systems

Offline Intelligence

Won SIH National Title by re-architecting a geospatial platform overnight for zero-connectivity environments.

5+Research Experiences
3Countries
1Publication
4National Awards
1000+Platform Users
4000Segmentation Dataset

Reading Influence Map

Annotated bibliography of papers that have significantly shaped my research taste and technical focus:

Shielding Applications from an Untrusted Cloud with Haven

Link

Andrew Baumann, Marcus Peinado, Galen HuntOSDI2014

Haven introduced the concept of shielding legacy, unmodified binaries inside Intel SGX enclaves. Reading this shaped my thinking on low-level context switches, virtual memory allocation overhead in secure enclaves, and why trusting the host OS is a fundamental design flaw for cloud deployments.

Distilling the Knowledge in a Neural Network

Link

Geoffrey Hinton, Oriol Vinyals, Jeff DeanNIPS Deep Learning Workshop2015

Hinton's foundational work on dark knowledge transfer. It directly influenced my wildlife tracking architecture, proving that heavy visual models like Grounding DINO and SAM can act as teacher networks to train lightweight edge-native student models without sacrificing structural localization accuracy.

Zero-Trust Architecture

Link

Alper Kerman, Oliver Borchert, Rose Rose, Scott RoseNIST Special Publication 800-2072020

This document formalizes that network location does not imply trust. It provided the security baseline for the GeoVision platform, driving my implementation of localized cryptographic authentication, strict RBAC bounds, and local-first data validation loops.

The CP-SAT Solver in Google OR-Tools

Link

Laurent Perron, Frédéric DidierOperations Research Proceedings2022

Reading the CP-SAT documentation and research notes changed how I approach scheduling. It showed me how to structure multi-variable operational fatigue limits into integer constraints ($x \in \{0, 1\}$) and apply spatial pruning heuristics to make NP-hard problems solvable in sub-second times.

Research Fields

Confidential Computing & TEEs

Hardware-enforced security primitives (Intel SGX, ARM TrustZone) to build secure computing boundaries in untrusted host environments.

Secure Edge AI

Running deep learning models locally under resource limits (e.g. dynamic INT8 quantization, NCNN, Raspberry Pi deployment) with cryptographic execution verification.

Trustworthy Distributed Systems

Architecting resilient, local-first protocols and decentralized computing frameworks for secure cross-node synchronization.

Cyber-Physical Systems & IoT

Securing operational hardware and command actuation loops (e.g. Arduino, ESP32 platforms) from spoofing, side-channel attacks, or host tampering.

Geospatial Edge Systems

Ingesting, processing, and visualising spatial telemetry pipelines locally without relying on active cloud networks.

Resource Optimization

Applying linear programming, integer constraints, and heuristic pruning tools (Google OR-Tools CP-SAT) to scale orchestration bounds.

Academic Journey

2018

National Sports Representative

Selected to represent the Patna Region as Vice-Captain in the XXIX National Handball Meet (Govt. of India), fostering physical discipline and team coordination leadership.

2021

BMS Mathematics Olympiad

Achieved Rank 4 in the regional Mathematics Olympiad organized by the Bihar Mathematical Society, demonstrating strong foundations in analytics and formal logical formulation.

2021

NAEST National Participant

Selected for the National Anveshika Experimental Physics Skill Test (NAEST) at IIT Kanpur, evaluated under the guidance of Prof. H. C. Verma.

2023

B.Tech in CSE (IoT & Cybersecurity)

Initiated formal study at Heritage Institute of Technology (current CGPA: 8.40), specializing in hardware-software intersections, local networking protocols, and cyber security fundamentals.

2024

TSEC Young Innovator Award

Built 'Jambavantha Irrigation', an IoT-based smart agricultural irrigation system that won the Tata Social Enterprise Challenge (TSEC) Young Innovator Award (top 0.05% out of 6,600+ entries).

2025

NJIT Backend Research (USA)

Completed research exchange at New Jersey Institute of Technology, implementing confidential computing vector pipelines on Intel SGX enclaves and Azure Confidential VMs.

2025

SIH Winner (National Title)

Led backend systems design for a team that won the Smart India Hackathon National Title (Top 0.002% nationally) under a Government R&D category.

2026

Government R&D Internship — UAV Video Intelligence

Built telemetry decoding, geolocation, and multi-modal detection pipelines for a drone video intelligence platform, including a U-Net segmentation model reaching 92.0% validation IoU.

2026

Loughborough Research Visitor (UK)

Collaborating as a Research Visitor at Loughborough University on AI-driven decision support engines using CP-SAT solvers and operations research scheduling methodologies.

2026

TRINETRA — Wildlife Conflict-Alert AI System

Leading a full detection, tracking, and risk-alert system for human-wildlife conflict, using PyTorch-based YOLO11 detection and ByteTrack multi-object tracking.

Future

Ph.D. Research Aspirations

Intending to pursue doctoral research on bridging low-level hardware security bounds (TEEs) with high-level distributed systems optimization for intelligent physical infrastructure.