I'm Ankit Sharma — a backend and platform engineer who builds systems that stay reliable, observable, and debuggable under production pressure. As an SMTS at Salesforce, I work across service architecture, API design, automation, and production reliability.
Outside work I ship open-source developer tools, mostly born from an incident I never want to repeat. Everything below is real; the framing is just more fun than a bullet list.
| Stat | Value |
|---|---|
| Class | Backend / Platform Engineer |
| Guild | Salesforce — Senior Member of Technical Staff |
| Region | India UTC+5:30 |
| Playstyle | Design for the incident, the operator, and the next maintainer |
| Passive | Ships the tool instead of writing the runbook |
| Current meta | Applied AI, with engineering discipline attached |
- Publish an observability toolkit that installs in one line
- Give coding agents local-first memory that survives the session
- Make production debugging boring
- Keep the streak alive
Live contribution graph. The ghosts are flaky tests.
The five encounters that keep respawning in production. This is the actual job.
| Boss | How it opens | Counterplay |
|---|---|---|
| Silent Latency Creep | p99 drifts for weeks; no alert ever fires | Percentile SLOs, request timelines, trace every hop |
| The Blocked Event Loop | CPU flat, memory flat, nothing moving | Watchdog on loop lag, move sync work off the path |
| Cascade of Retries | One slow dependency becomes an outage | Timeouts, budgets, circuit breakers, backpressure |
| The Contract Break | Nothing errored, everything is subtly wrong | Explicit API contracts, typed edges, contract tests |
| The Unexplained Deploy | It broke at 02:00 and nobody knows what shipped | Small commits, real messages, searchable history |
Win condition isn't "it works." It's "it keeps working, explains itself when it doesn't, and the next engineer can operate it."
🕹️ PRESS START — a playable incident. Every ending is a real tool.
03:12. The pager goes off. Checkout
p99just went from180msto8s. Error rate is climbing. Nobody else is awake. Choose your opening move.
▸ 1. Roll the pods. It usually works.
Latency drops. You go back to bed.
04:40. The pager fires again. The queue is deeper this time.
▸ Roll them again.
💀 GAME OVER — "The Restart Loop."
Three restarts, zero knowledge. Tomorrow's review will ask what changed, and the evidence went out with the last pod. Restarting resets the symptom and deletes the crime scene.
Continue? → node-actuator-lite gives you health, env, and Prometheus endpoints so a pod can tell you why it's sick before you kill it.
▸ Grab a heap snapshot first, then roll.
✅ RECOVERED.
Heap grows linearly with open connections and never comes back down. That's not load — that's a leak. You have a real bug and a real fix.
Loot → node-observability-lite — actuator endpoints, request timelines, and an event-loop watchdog in one line.
▸ 2. Open the dashboard first.
Good instinct. What's on it?
▸ CPU flat. Memory flat. Requests piling up in the queue.
💥 CRITICAL HIT.
Nothing is busy and nothing is moving — the classic signature of a blocked event loop. One synchronous call on the hot path is holding the whole process hostage. Everything behind it is just waiting in line.
Loot → node-eventloop-watchdog — catches loop stalls before they become incidents.
▸ There is no dashboard for this service.
💀 GAME OVER — "Debugging Blind."
You're now reading raw logs at 3 A.M. and guessing. The outage isn't the incident; the missing telemetry is. This one was lost at design time.
Continue? → node-request-trace — per-request tracing and performance visualisation you should have added last quarter.
▸ 3. Ask what shipped in the last 24 hours.
The right question. You open the log.
▸ It's a wall of "Merge pull request #…"
🏆 TRUE ENDING — "Root Cause."
200 commits, no story. You stop scrolling and start querying: filter by path, by author, by the window the graphs turned. One config default flipped, buried in a rename. Ninety seconds, not ninety minutes.
Loot → git-history-ui — local-first Git investigation with portable reports and PR impact analysis.
▸ …wait, what's this commit at the bottom?
🥚 SECRET FOUND. You read all the way down here. Genuinely, respect.
Take the whole toolbox: every package I've shipped. Or say hi — I like people who read the stack trace to the end.
▸ One deploy. 400 files. Message: "small fix".
😐 ENDING — "Small Fix."
It was not a small fix. It is never a small fix.
Loot → roastcode — an AI CLI that roasts code, commits, and diffs, so the diff gets judged before production does.
CORE — languages I build in
BACKEND + AI — the main branch
DATA + INFRA — where it actually runs
SIDE BRANCH — enough frontend to ship the whole thing
|
Applied AI product Conversation analysis, attraction scoring, ghosting-risk signals, and context-aware reply suggestions. 
|
AI knowledge system Personalized guidance across sacred texts, traditions, languages, and reflective modes.
|
Interactive learning tool Explore algorithms, data structures, coding games, and Git through interactive visualizations.
|
Shipped and installable. Rarity is graded on real download volume — the counters are live.
| Item | What it does | Rarity | Version | Pulls |
|---|---|---|---|---|
| git-history-ui | Local-first Git investigation, portable reports, PR impact automation |  |
 |
 |
| node-actuator-lite | Actuator-style health, environment, and Prometheus endpoints for Node.js |  |
 |
 |
| jambavan | Local-first MCP memory and code intelligence for coding agents | |
 |
 |
| meme-as-a-service | Generate and serve memes programmatically |  |
 |
 |
| node-request-trace | Request tracing and performance visualization for Node.js APIs | |
 |
 |
| pravaah | Production-grade Excel, CSV, and JSONL pipelines for Node.js | |
 |
 |
| node-eventloop-watchdog | Detect event-loop stalls before they turn into production incidents |  |
 |
 |
| node-observability-lite | One-line actuator endpoints, request timelines, and event-loop watchdog | |
 |
 |
| readme-cinema | Transform READMEs into cinematic terminal experiences | |
 |
 |
| roastcode | AI-powered CLI that roasts code, commits, and diffs | |
 |
 |
| eli5 | AI-powered "explain like I'm five" annotations for Java docs |  |
 |
 |
Open to good problems, open-source collaboration, and anyone who has also debugged something at 3 A.M.
Built one production system at a time by @beingmartinbmc. The run continues.