Software Engineer at Cisco | Backend & Distributed Systems | Go, Java, PostgreSQL, Kubernetes
Bengaluru, India. I build and debug backend services, messaging workflows, durable state, and observable Kubernetes systems. My public engineering work focuses on failure recovery, safe retries, resource lifecycle, and reproducible tests.
Live portfolio · LinkedIn · Email
Earlier GitHub work: @ajaykr2712.
- Released the Distributed Scale Validation Lab v0.1.0, an experimental durable asynchronous Go backend with separate API, outbox dispatcher and workers, RabbitMQ and PostgreSQL. Twelve real-service trials and recovery receipts document single-host limits and observation overhead. The earlier 10,000-job result remains an in-memory functional baseline, not external-path performance.
- Contributing reliability fixes to NATS and Prometheus through reproductions, regression tests, and maintainer review.
- Maintaining the Fault-Tolerant Transformer Lab as a second case study in deterministic checkpoints and restart verification.
- Merged: NATS Python PR #1043 preserves explicit passwords alongside signed CONNECT credentials. Maintainer-reviewed and merged 2026-10-01 21:36:34 UTC.
- Merged: Prometheus PR #19882 restores the active timestamp of previously firing alerts with short
fordurations so they resume firing on the next scheduled evaluation. Maintainer-reviewed and merged 2026-10-05 10:20:20 UTC; pending-alert restoration and grace-period behavior remain unchanged. - Open, checks passing: NATS Python PR #1051 cleans up temporary KV watchers and closes a cancellation/closed-state race. All 27 checks passed on 2026-10-08 07:03:33 UTC; normal maintainer review is pending. It is not counted as merged work.
- Closed, not merged: TorchTitan PR #4864 was superseded by upstream RL CI work. It is not counted as an accepted contribution.
- Public pull requests track work submitted for external review.
Go Java PostgreSQL RabbitMQ NATS Kubernetes Prometheus OpenTelemetry Python
| Project | What it proves | Status |
|---|---|---|
| Distributed Scale Validation Lab | Durable acceptance, idempotency, outbox publication, commit-before-ack, real-service crash/restart tests and raw worker trials | Experimental v0.1.0 released; single-host lab |
| Merged infrastructure contributions | NATS credential serialization and Prometheus alert-state restoration, accepted under normal upstream review | Merged |
| Fault-Tolerant Transformer Lab | Gradient checks, atomic checkpoints, deterministic restart equivalence | Runnable CPU recovery lab |
| Engineering Portfolio (source) | Case studies with exact code, evidence, and limitations | Live |
Every featured claim should resolve to code, a test, a reproducible command, or a versioned artifact. Planned work is labelled as planned rather than shipped.
90-second recovery test replay · Exact output receipt. This is actual agent-executed terminal output, not a personal presentation or production evidence.