Santosh Parsa
Senior Staff Systems & Infrastructure Engineer with 13+ years architecting, automating, and securing mission-critical Linux platforms, distributed cloud systems, and AI-assisted reliability toolchains.
[Systems & Network] Linux Kernel, TCP/IP, DNS, eBPF Profiling
[Cloud & Containers] AWS, GCP, Kubernetes, Terraform
[Middleware & Data] Apache, Tomcat, Nginx, HAProxy, MySQL, Postgres
Featured Projects
Open-source tools and autonomous telemetry systems
DAG-Ops: Deterministic AI Engine
Dependency-free distributed DAG execution engine applying workflow graph principles and Kahn's topological sort to enterprise LLM pipelines, delivering 42% faster triage and 48% lower token costs with isolated fault boundaries.
Git Systems Architecture & Conflict Engine
Interactive systems engineering platform dissecting Git's content-addressable Merkle DAG, 4-zone state machine, and live 3-way Lowest Common Ancestor (LCA) merge resolver.
ForgeOps AI
A safety-first autonomous reliability platform acting as an unprivileged SRE agent over infrastructure via Model Context Protocol (MCP) with prompt-injection neutralization.
FastMCP Linux Observability Server
Lightweight Model Context Protocol microservice exposing real-time CPU, RAM, and storage health metrics over Server-Sent Events (SSE) and local stdio with zero-root diagnostic access.
Enterprise Mail & Ticket Intelligence
Automated incident triage and email ingestion engine with local Ollama routing for confidential PII and frontier models for escalations, cutting API overhead by 85%.
Engineering Labs
Reproducible benchmarks, kernel profiling, and SRE experiments
DAG-Ops: DAG vs. Monolithic LLM Simulator
Live visual state machine and telemetry scoreboard comparing monolithic prompt failure crashes against Kahn's topological sort and isolated per-node retries.
How Git Works Under the Hood: State Simulator & Conflict Engine
Interactive 4-zone state machine simulator, live 3-way merge conflict editor, content-addressable DAG object database, and 10 core engineering deep dives.
Linux Kernel TCP Backlog & Socket Diagnostics
Isolating listen backlog overflows versus accept queue saturation under traffic bursts using ss -lnt, netstat -s, and sysctl tuning.
Multi-Window Multi-Burn-Rate SLO Alerting
Codifying 14.4x 1h and 6x 6h burn-rate alerts against 99.9% SLOs in Prometheus & Alertmanager to eliminate alert fatigue by 65%.
Thread Contention & JVM Profiling on Linux
Mapping high CPU threads from top -H to hexadecimal thread dumps (jstack) and tracing locking bottlenecks under production loads.