Hyderabad, India · IIIT-H CSE ’28
Ayush Agarwal
Software engineer · systems, infrastructure & the hard parts
B.Tech CSE at IIIT Hyderabad (CGPA 9.44/10), currently a software engineer intern at ConnectMyWorld. I like building foundational software from scratch: a distributed file system with automatic failover, kernel demand paging and CPU schedulers, a reliable transport protocol over raw UDP, and production on-device ML used by 650+ people daily. Every project below has an interactive deep dive. Open one, drop packets, crash servers, and watch the system recover.
01 · Experience
Where I’ve shipped to production
Two internships at the same company, and they asked me back.
Software Engineer Intern
ConnectMyWorld May 2026 – Present- Prevented proxy attendance by designing and shipping an on-device, offline-first face-verification system across the company’s Android and iOS apps, now deployed and actively used by 650+ users.
- Built the full ML pipeline locally on the phone: Google ML Kit detection, ArcFace-5 alignment, MobileFaceNet embeddings, a multi-template matching scheme, tuned thresholds, and an active anti-spoofing challenge. No face image or embedding ever leaves the device.
Software Engineer
NJSecure Jan 2026 – Apr 2026- Led a 5-person team building and shipping Orbit-Node, a security-orchestration platform for a bank network, for NJSecure (a US client) as our software-systems capstone at IIIT Hyderabad.
- Owned the architecture and built the core myself: the AI reasoning, the Gemini MCP server, and the orchestrator over a Neo4j asset graph, with Presidio anonymizing sensitive data before it reaches the model. Also designed and contributed roughly half of the Wazuh, Neo4j, and Presidio modules. Delivered end-to-end, earning an A grade and client recognition.
Software Engineer Intern
ConnectMyWorld Feb 2025 – Jul 2025- Built the iOS version of the company’s native-Java Android app in Flutter/Dart (with native Swift), solving the core challenge of keeping a continuous field-employee tracking app running persistently as an iOS background process.
- Engineered reliable collection and offline sync with zero data loss of location and form data, hardened the app against adversarial users with tamper-resistant logic; validated through extensive case testing and deployed to production, where it is actively used by 1000+ real users.
02 · Projects
Systems, built from scratch
Each card opens an interactive deep dive: a live, clickable model of the system. Drop packets mid-flight, crash the primary server, trigger page faults, approve an AI-proposed mitigation. The demos run the same logic the write-ups describe.
Orbit-Node: Security Orchestration
A bank security-orchestration platform built for NJSecure, a US client: alerts are enriched against a Neo4j asset graph, reasoned over by an AI agent, and executed by the Wazuh SIEM, but only after a human analyst approves. Led a 5-person team.
Interactive deep dive →CoWrite: Distributed File System
A collaborative document store with Google-Docs semantics over raw TCP: sentence-level write locking, per-user ACLs, primary/backup replication with automatic failover, and checkpoint/undo rollback.
Interactive deep dive →xv6 Kernel Enhancements
Extended the xv6-riscv teaching kernel with demand paging (lazy allocation, page-fault handling, FIFO swap eviction) and four pluggable CPU schedulers (FCFS, RR, CFS, MLFQ) with benchmarking instrumentation.
Interactive deep dive →SHAM: Reliable Transport over UDP
TCP-style reliability built from scratch on best-effort UDP: 3-way handshake, sliding-window flow control, cumulative ACKs, RTO retransmission, and in-order reassembly, all validated under deliberately injected packet loss.
Interactive deep dive →On-Device Face Verification
A fully offline face-verification gate for a field-attendance app: ML Kit detection, ArcFace alignment, MobileFaceNet embeddings, multi-template matching, and active liveness. In production with 650+ users.
Interactive deep dive →C-Shell: a Unix shell
A custom Unix shell in C: command parsing with pipes and redirection, foreground and background execution with job tracking, built-ins, and signal handling.
View on GitHub →C-Shark: Packet Sniffer
A command-line network analyzer on libpcap: interface discovery, live capture
with filters, and packet inspection down to individual protocol fields.
Threaded Bakery
A multithreaded bakery simulation modelling customer arrivals, seating, baking, and payment under strict synchronization constraints, with deterministic timestamped logs.
View on GitHub →Audio-Guided Microbot
A voice-controlled microbot on ESP32-S3 with on-device ML: TensorFlow Lite Micro recognizes spoken commands in real time, with BLE manual override and a Python monitor.
View on GitHub →03 · Skills
Tools I actually use
Everything here appears in at least one project above.
Languages
Systems & Concurrency
AI / ML
Backend & Infrastructure
04 · Education & Achievements
Background
International Institute of Information Technology, Hyderabad
Relevant coursework: Data Structures & Algorithms, Computer Systems Organization, Algorithm Analysis & Design, Database Systems, Operating Systems & Networks, Information Security, Machine Learning, Software Design & Analysis.
Achievements
- Dean’s List, 4 out of 4 semesters, awarded to the top 10% of the batch.
- Problem setter (CTF challenges), Hacking Club IIIT-H, for 1.5+ years.
- 2nd place (Open), Deccan CTF.
- JEE: AIR 836 (Main), AIR 1630 (Advanced).
05 · Contact
Let’s talk
Looking for SWE internship opportunities in systems, infrastructure, and backend engineering. The fastest way to reach me is email.