Systems researcher with a consistent focus on one question across all my work:

Where should computation happen, and how does that decision affect system behavior?

That question has driven me from building distributed mobile platforms, to benchmarking edge-vs-cloud latency with real IoT hardware deployed across three live classrooms (published, KCI-indexed), to evaluating on-device ML inference as an architectural decision in mobile systems. Three peer-reviewed publications across embedded systems, fog computing, and distributed learning environments, all grounded in deployed systems and real measurements.

I am currently a research intern at SSLab (Intelligent Software Systems Lab), Kyungpook National University, working on Kubernetes scheduling, edge computing middleware, and serverless infrastructure, with plans to continue into KNU's master's program in Computer Science and Engineering starting Spring 2027.

SSLab, Kyungpook National University · Sep 2026 – Present

Working on systems research at the intersection of container orchestration and cloud infrastructure reliability. Current project studies how Kubernetes pod relocation disrupts running workloads, measuring how disruption time varies with target-node state (image cache, resource pressure, network conditions), and building a predictive model to inform better placement decisions than Kubernetes' default heuristics. Targeting a submission to a systems conference (CCGrid or ACM SAC) in the 2026–2027 cycle.

This builds on a broader lab research line spanning edge computing middleware, serverless computing, and RL-based scheduling for distributed infrastructure.

From undergraduate research, 2025–2026

[1] Gyawali, A., Al-Absi, A. A., Abukhalifeh, A. N., Al-Absi, M. A. "EcoSense: Edge vs. Cloud Computation Placement for Real-Time IAQ Monitoring in University Campus Environments: A Pipeline Latency Analysis." International Journal of Advanced Smart Convergence (IJASC), Vol. 15, No. 2, pp. 282–296, June 2026. KCI-indexed. 🔗 ResearchGate · doi.org/10.7236/IJASC.2026.15.2.282

First-authored. 5,317 readings across 3 real classrooms. Edge median 715 µs vs. cloud 6 ms, full statistical separation (Mann-Whitney U=0, Cliff's δ=1.0). Dataset released on Zenodo.

[2] Gyawali, A., Bhandari, K. S., Al-Absi, A. A. "On-Device LLM Reasoning for IoT Anomaly Detection in Fog Computing Environments." Environment-Behaviour Proceedings Journal (E-BPJ), 11(37), pp. 165–172, June 2026. DOAJ-indexed; WoS indexing under evaluation. 🔗 doi.org/10.21834/e-bpj.v11i37.7945 · Oral presenter, AicE-Bs2026Sokcho · Overall rank 7/36 · Best Paper rank 5/26

[3] Gyawali, A., Al-Absi, A. A., Al-Athwari, B. "Physical Learning Environments and AI-Powered Personalized Learning in Higher Education: A Systematic Literature Review." Environment-Behaviour Proceedings Journal (E-BPJ), 11(37), pp. 30–46, June 2026. DOAJ-indexed; WoS indexing under evaluation. 🔗 doi.org/10.21834/e-bpj.v11i37.7897 · Oral presenter, AicE-Bs2026Sokcho · Overall rank 15/36 · Best Paper rank 9/26

[4] Gyawali, A., Al-Athwari, B. "Deploying a University Social Platform on Free-Tier Serverless Infrastructure: Performance Characterisation and Cold-Start Source Isolation." ResearchGate, 2026. 🔗 View on ResearchGate

[5] Gyawali, A. "From PWA to Native: A Case Study in Migrating a React/TypeScript Application to Flutter." ResearchGate, 2026. 🔗 View on ResearchGate

[6] Gyawali, A. "Comparative Evaluation of Lightweight Face Detection Models Under Programmatic Degradation: A Controlled Pilot Study with WIDER FACE Benchmark Validation." ResearchGate, June 2026. 🔗 doi.org/10.13140/RG.2.2.29797.36328

🏆 Best Poster Award, Smart Computing & AI Implementation Category KDU Global Research & Innovation Fair, Spring 2026 · Graduate-Undergraduate Research Programme (GURP) Paper selected for publication in IJASC (KCI-indexed)

Systems & Infrastructure

Languages

Mobile & Embedded

Web & Backend

Tools

Undergraduate research project under KDU's Graduate-Undergraduate Research Programme (GURP), Spring 2026. ESP32 WROOM-32D with four sensors (DHT22, MH-Z19C, MQ-135, DFR0034) connected to a Flutter Android dashboard via BLE, with opportunistic Firebase Realtime Database sync.

Research question: does local edge processing produce lower end-to-end latency than cloud offloading under real campus network conditions? Yes, 8.4× faster: edge median 715 µs vs. cloud 6 ms across 5,317 readings in three classrooms. Distributions fully separated (Mann-Whitney U=0, Cliff's δ=1.0). Published in IJASC (KCI-indexed), June 2026. Dataset on Zenodo. Won Best Poster Award at KDU Research & Innovation Fair.

🔗 View Repository · 📄 Publication [1] · 📦 Dataset on Zenodo

Full-stack campus social platform built after observing international students at KDU struggling with Korean-only interfaces. Features campus feed, file sharing, group discussions, and a pseudo-anonymous cookie-based identity system with no registration required.

Key engineering contributions: httpOnly cookie-based anonymous persistent identity via UUID, and an upload intent validation workflow preventing presigned URL path injection. Three-tier architecture: React 18 + TanStack Query frontend, Express.js backend (35 REST endpoints), Neon PostgreSQL with Drizzle ORM and Multer disk storage, deployed on Render.

Serverless cold-start characterisation on live infrastructure: container wake-up isolated at mean 52.4 s vs. database resume at 621 ms, two orders of magnitude apart. Validated under concurrent load: zero errors across 10,584 requests at 100 VUs, p95 < 800 ms.

🔗 View Repository · 📄 Preprint [4]

Native mobile rebuild of LevelUp. Evaluates on-device versus cloud inference as a systems architecture decision: where should ML computation run in a resource-constrained mobile environment? On-device ML Kit image labelling for habit photo verification, so no image leaves the device. Offline-first architecture via SyncService and NetInfo with conflict-aware background sync to Firebase. Anti-cheat telemetry engine with 7 heuristic rules (Haversine GPS, accelerometer variance, step-to-distance ratio, speed variance). Jest test suite covering XP calculation, streak risk, and anti-cheat validation.

🔗 View Repository

Gamified habit tracker built around RPG progression: XP, levelling, stat attributes, streaks, and a credit economy. Hit fundamental web limitations (no camera, no background processing, no push notifications) that motivated a full native mobile rebuild as System Override.

🔗 View Repository · 🌐 Live Demo

Freelance project: production website for a beauty training institute in Kathmandu, Nepal. SEO-optimised (semantic HTML, core web vitals), localization-ready architecture (English/Nepali dictionary pattern), animated responsive components.

🌐 Live Site

Role-based academic management system with separate professor and student dashboards, class scheduling, grading, and file management. Built after observing administrative fragmentation at KDU. CSRF protection and policy-layer access control throughout.

🔗 View Repository