Siddhant Kumar Upmanyu
Senior Systems Architect
4+ years of full ownership over backend and cloud infrastructure at an IoT smart home company. I design at the protocol level when needed — custom binary over UDP, kernel-level traffic routing, nftables packet reflection. I publish the tools I build: a Kotlin microservice framework on Maven Central, a Go assertion library, a Helm CLI in Rust, a custom Linux distro. TDD is non-negotiable across every language and every project.
Experience
Architecture & API Design
- Own the API contract suite for the entire platform — 30+ specs covering device provisioning, scene/rule automation, OTA, smart lock integrations, and real-time device status. App and firmware teams build against these before a line of code is written.
- Lead technical specification for the unified backend and cloud infrastructure stack, security standards, and data contracts.
IoT Protocol & Real-time Systems
- Designed and implemented the custom binary protocol (UDP/TCP) between IoT hubs and cloud — packet handling, NAT hole punching, hub online/offline state machine, connection tracking, race condition handling.
- Pushed network work to the kernel when needed: nftables UDP echo server for NAT traversal (packet reflection without userspace), IP rate limiting via per-source dynamic sets at prerouting priority.
- Tuned kernel parameters for high-throughput IoT workloads: socket buffers, IP forwarding, connection tracking table size.
Platform & Automation Engine
- Built the scene and rule automation engine: multi-hub scenes, nested fragment handling for large payloads, scene sync, IFTTT-style cloud rules, scheduling.
- Built the device provisioning system from scratch (extracted from monolith): hub setup, node validation, WiFi flow rewrite, replacement flow for failed nodes.
- Integrated Google Home, Alexa, and Yale smart locks (cloud-to-cloud) into the platform.
- Built a feature flags system for controlled rollouts.
- Introduced Consumer-Driven Contract Testing (Pact) for firmware delivery APIs — full CI pipeline: publish pact → can-i-deploy → release → mark deployed.
Performance Engineering
- ClickHouse migration: rewrote core ingestion pipeline in Go, achieved 95%+ storage reduction (90 GB → 3.7 GB, 25 GB → 780 MB) via schema design and specialized codecs.
- JVM profiling with async-profiler and flame graphs — eliminated CPU spikes from per-packet object allocations using ThreadLocal; tuned GC for throughput.
- Zero-downtime blue-green deploys via iptables: flushed conntrack for immediate cut-over, applied rules to both PREROUTING and OUTPUT chains for internal routing.
- Implemented structured ThreadPool error handling after debugging silently swallowed exceptions in async paths (no logs, no traces — the bug that inspired the blog post).
Infrastructure & Reliability
- Migrated disk-based logging to Loki; built Prometheus + Grafana monitoring — significantly reduced MTTR.
- Architected EKS migration using Terraform; evaluated HashiCorp Nomad first, selected EKS.
- Implemented TLS client authentication with a custom CA and certificate revocation list (CRL) in Go.
- Maintained production Linux systems: kernel tuning, live Debian/Ubuntu upgrades,
tcpdump,strace,sar,iotop. - Built and maintained GitHub Actions CI/CD pipelines for automated testing, building, and deployment.
Codebase & Quality
- Migrated core monolith (Spring 4 / Java) to Spring Boot 2.7+ / Kotlin — 100% TDD throughout to prevent regressions.
- Decoupled monolith into gRPC microservices.
- Led technical recruitment: designed coding challenges, conducted interviews.
Open Source
Microservice framework on Helidon SE (Nima) + Project Loom. Custom compile-time DI via KSP — no runtime reflection. Three-tier testing: unit → in-process integration server → Docker E2E via Testcontainers. Gradle plugin that wires the full build pipeline. Published at dev.sku20.stopgap:*:2.8.0.
Type-safe assertion library with generics. Fluent API, chainable Not(), custom matchers, zero dependencies. v1.0.0.
Helm Release Helper — reads a YAML release declaration, invokes helm upgrade --install. Diff mode, --atomic rollback, chart versioning. Installable via cargo install.
Minimal Linux distribution based on Void Linux, built for servers and recovery disks. Ships .img.gz and .qcow2 images via GitHub Actions releases.
OOP and London-style TDD in pure C. Polymorphism via interface structs, heap-allocated objects with constructors. v5.0.0, CI.
Publications
Redis-based architecture for tracking IoT device online/offline status across stateless distributed backends — sorted sets scored by expiry deadlines, throttled batch writes, asymmetric fault-tolerance guarantees ensuring offline transitions are never lost. Derives worst-case detection latency bounds. v1.3.0.
Technical Writing
Writing at eglu.tech about production engineering, JVM internals, and systems design.
- Zero-Downtime Deployments with Iptables
- Java Exceptions Swallowed: The ThreadPool Trap
- ThreadLocal Optimizations and Project Loom
- A Low-Level UDP Echo Server for NAT Traversal
- What is Rate Limiting?
- Optimizing Hex Formatting
- Avoid — A Void Linux Distribution
Skills
- LanguagesKotlin, Java, Go, Rust, Zig, C, Shell
- Frameworks & ProtocolsSpring Boot, Helidon SE, gRPC, Protocol Buffers
- InfrastructureKubernetes (EKS), Terraform, Docker, GitHub Actions, Linux (Debian/Ubuntu)
- DatabasesClickHouse, MongoDB, PostgreSQL, Redis
- ObservabilityLoki, Prometheus, Grafana, async-profiler, JMC, VisualVM, Flame Graphs
- TestingTDD (London style), Pact (Consumer-Driven Contracts), Playwright, Testcontainers
- NetworkingTCP/UDP socket programming, iptables, nftables, NAT traversal