I'm a CS undergrad at SIT Pune, but my stack doesn't know that. Kafka, Kubernetes, distributed transactions, observability stacks — I build backend systems the way they're supposed to be built. Then I benchmark them.
Go |
TypeScript |
JavaScript |
Python |
Java |
Rust |
Node.js |
Gin |
FastAPI |
GraphQL |
gRPC |
PostgreSQL |
MongoDB |
Redis |
Kafka |
RabbitMQ |
ClickHouse |
Neo4j |
AWS |
Docker |
Kubernetes |
Terraform |
Helm |
ArgoCD |
NGINX |
Kong |
Cloudflare |
Prometheus |
Grafana |
OpenTelemetry |
Datadog |
Actions |
Linux |
React |
Next.js |
Tailwind |
Jest |
k6 |
kprobe — Kernel Observability for Financial Systems
Rust Aya eBPF Go Python Kafka Neo4j ClickHouse OpenTelemetry Jaeger
Rust/Aya eBPF probe attaching to kernel tracepoints at runtime — captures TCP RTT, CPU scheduling, and syscall latency at nanosecond resolution with zero application instrumentation. A Go/Kafka stream processor windows events into 100ms frames, builds a directed causal graph by PID/transaction co-occurrence, and classifies failure chains into Neo4j for root cause attribution. Isolation Forest anomaly detection over ClickHouse time-series triggers an LLM-powered RCA pipeline (Gemini API) that generates incident reports from the causal subgraph.
ESX — Stock Exchange Infrastructure
Go Kafka gRPC PostgreSQL Redis AWS (EKS, MSK) Kubernetes
Production-grade securities exchange — 8 microservices (order gateway, settlement, matching engine, clearing house) over Kafka and gRPC. Order gateway sustains 6,000+ RPS at sub-40ms latency via gRPC streaming and a buffer-and-drain pattern (500k-capacity buffered channels), surviving 300% load spikes. Atomic DvP settlement with an idempotent double-entry ledger — all movements net to zero.
Hatch — Self-hosted Cloud Deployment Platform
Go Gin AWS (ECS, Fargate) RabbitMQ Redis PostgreSQL Terraform Docker
Self-hosted Render/Railway alternative — independent Go microservices for cloning, building, and deploying containers on ECS Fargate, communicating via RabbitMQ with real-time log streaming through Redis pub/sub over WebSockets. 3,500 RPS at p99 < 75ms with zero failed requests (ApacheBench), sustaining 716 req/s at p95 < 9ms under 50 concurrent VUs (k6).
|
|
|
|
|