I build production AI systems and correctness-minded infrastructure. At Fulcrum I ship agentic RAG, event-driven analytics, and internal AI assistants. Outside work I write Rust systems — a simulation-tested Raft KV, a zero-instrumentation eBPF agent, and a multi-tenant LLM serving gateway. Always open to bold projects — let’s build what’s next.
CO
DING.
LIN
.KS
{ }
TECH
STACK
Languages:
AI/ML:
Backend:
Infrastructure:
Explore my latest work and side projects
View All ProjectsChronos
- Built a simulation-first Raft KV store in Rust: shared step(node, event) → effects protocol under a deterministic simulator (owned time/network/disk/RNG) and a single-node WAL node interpreter. - Implemented Raft election + replication on static 3/5 node clusters; Get/Put via the log with idempotency; five executable safety checkers (election/log safety, persistence, linearizability) validated across 1,000+ fuzzed seeds clean in CI. - Built swarm fuzz with seeded fault injection, byte-for-byte reproducible replay, SHA-256 digests, and schedule minify — failure artifacts you can re-run, not flakes.
eBPF Observability Agent
- Built a zero-instrumentation eBPF agent that reconstructs HTTP/1.1, HTTP/2, and gRPC request latency plus a live service map purely from kernel syscalls and OpenSSL uprobes — no application code changes. - Deployed as a Kubernetes DaemonSet on k3s with OTLP export, sticky-sampling for overload survival, and on-CPU stack profiling joined to slow spans. - Characterized the limits of full-fidelity tracing under peak load, finding ~87% single-core overhead versus an original <2% target and motivating adaptive sampling and load-shedding strategies.
Atlas
- Shipped an OpenAI-compatible multi-tenant inference gateway with tenant auth/quotas and three routing strategies (round-robin, least-load, prefix-aware), tested on RTX 3050 4GB and Colab/Kaggle T4 GPUs. - Measured a 5.5× p50 latency improvement when replacing naive Hugging Face serving with vLLM under concurrency. - Found that sticky prefix-aware routing caused a 2× TTFT regression in simulation and a smaller ~1.17× regression on dual live vLLM instances; implemented a load-gate mitigation.
Axon (Observatory)
- Developed comprehensive monitoring platform tracking API performance across distributed systems with real-time dashboard and alerting. - Built intelligent alerting system with BullMQ background worker and email notifications handling monitoring sessions. - Created npm monitoring agent with high-precision timings and efficient batching for seamless Express.js integration.