Senior Backend Engineer/Ruby on Rails & Elixir/Phoenix
I build and run the backend systems that keep US home-care agencies paid and Medicaid-compliant — multi-tenant healthcare SaaS processing ~270K visits/day across 27 agencies in 8 states. 10+ years backend, 5+ years fully remote on US time zones. Rails and Elixir/Phoenix daily, AI-augmented with Claude Code and Cursor.
grep -i "incident" postmortems/
Diagnosis is the job. These are the three I'd tell you about in an interview.
Every alternate Monday — payroll day — the platform crashed under biweekly visit-export load. Predictable, recurring, business-critical.
I led a layered investigation that resolved five stacked bottlenecks, each one exposing the next: staging and production sharing node pools; every tenant's export job firing at the same instant (staggered them — smoothed load and cut cost); pods too under-threaded to ever cross the HPA's 80% CPU threshold, so autoscaling silently never fired; DB connections exhausted (recalculated as max pods × workers × threads); and residual slowness from an upstream VM-hosted service with a too-narrow Passenger pool.
No recurrence since. Resource utilization healthy, exports fast, payroll Mondays boring.
During our GitHub → GitHub Enterprise migration, I spotted a silent failure during a routine deploy: one shared access token, used by every microservice to fetch client configuration, had been invalidated.
The trap: running servers looked perfectly healthy — they were serving cached config. Any restart would re-fetch with the dead token and take the site down. Staging and production sites began failing one by one across tenants.
There was no shortcut — every microservice needed redeployment with the new token. I took ownership with my engineering manager, coordinated fixes across every team and server, and supported the rollout in real time. All tenants stable within one day. Early identification was the difference between a one-day fix and an outage nobody could have recovered quickly.
Loading Dock's twice-weekly production batch was taking 14–16 hours to import a week of visit history — agencies waited nearly a full day on payroll day.
I traced it to redundant inter-service lookups — the same Employee and Consumer records re-fetched on every visit, dozens of times per pair per import. Fix: a job-scoped Redis cache keyed on the import file, cleared on completion.
The judgment call: I chose deduplication over throwing workers at it deliberately — parallel processing risked duplicate timesheets with downstream payroll impact. Correctness constrained the solution space.
tail -f career.log
GlobalLogic Inc · Senior Software Engineer (Consultant)
Contracting for HHAeXchange (US). Technical lead representing HHAeXchange within GlobalLogic — authored the hiring JD, conducted ~20 technical interviews, hired 4 engineers (all still with the company), and lead a cross-functional team of 5. Stabilized an org-wide CI/CD migration to GitLab CI + GitHub Enterprise and drive API contract unification across three versioned monoliths.
NSR Information Systems · Senior Software Engineer
HHAeXchange / Annkissam. Delivered full-stack features across EVV compliance, fiscal management, and onboarding platforms; improved performance on large-scale multi-tenant apps using Datadog and Scout.
RayMn & BR Technologies · Senior Software Engineer
Improved average response time 20% through query optimization, caching, and background jobs. Integrated Braintree and Billplz payment gateways in production. Mentored junior developers — 30% reduction in production bugs.
Intelliswift Software · Software Engineer
Delivered client-facing Rails applications with distributed remote teams. Integrated Stripe for ParkSmart, a parking-management product still live in production today.
RayMn & BR Technologies · Rails Developer
Initiated and led Rails projects from scratch — admin panels, REST APIs for mobile apps, deployment pipelines. Integrated PayPal for Planiversity, still live in production today.
ls -la services/
Production systems, sorted by lines of code added. All built and run for HHAeXchange-affiliated healthcare SaaS platforms.
Timesheets module end-to-end (6,993+ lines), no local persistence — sourced live through upstream APIs. Driving API contract unification across three backend versions.
Multi-state EVV compliance platform: 27 agencies, ~10K visits/day per tenant, 8 states. Shifts API v2 with batch entry and exception codes — 4,500+ line mobile-integration feature.
Healthcare EVV integration platform. Job-scoped Redis cache cut the twice-weekly ingestion batch from 14–16 hrs to 2–3 hrs. External API layer for third-party clients.
Cache V2 API powering offline mobile support — parallel async fetch across modules, stitched into one fast response. Multistate financial-entry compliance support.
GraphQL API for ShiftField queries with full VCR test coverage. Diagnosed and fixed a production HTTP bottleneck by tuning Finch connection pooling.
In-house Elixir deployment tool bridging DevOps and engineering. Live Phoenix LiveView monitoring; removed ~1,400 lines of redundant node-affinity config.
HR onboarding platform. I-9 form end-to-end with multi-preparer support and a scalable, configurable per-client e-signature template system.
Multi-tenant IVR visit-collection platform ingesting via Amazon SQS. Entire UI runs on real-time Phoenix LiveView, with tenant-level configuration isolation.
Personal · Open source · AI
Google Maps/Places/Search API wrappers, a static map image generator, and cross-environment DB cloning rake tasks. Deprecated gems maintained with proper migration notices — open source, responsibly.
Fully local retrieval-augmented pipeline over a personal Obsidian vault — no cloud APIs, no data leaving the machine. Three-tier chunking, ChromaDB retrieval, Typer CLI.
Keyboard-driven terminal chat client for OpenAI models. Streaming token output, mid-stream cancellation, per-conversation system prompts, model switching.
A project-scoped note-taking system for Neovim — fuzzy search, daily notes, and Telescope integration. Written to fix my own workflow, published for anyone's.
Shell-based automatic time tracking on TimeWarrior — starts and stops based on tmux session, git branch, and directory, with ticket caching and per-terminal isolation.
cat workflow.md
AI-augmented, not AI-dependent
Claude Code and Cursor are in my daily loop, with agent autonomy boundaries explicitly configured — no auto-commit, no auto-push. The agent proposes; I review and own what ships.
Terminal-first
Neovim, tmux, Lazygit, and tooling I've written for myself — a published Neovim plugin, an automatic time tracker, a local RAG pipeline over my own notes. If a workflow annoys me twice, I build something.
ls posts/ --sort=relevance
Zero-Downtime PostgreSQL Major Version Upgrades in Containers: The Problem Nobody Talks About
Why in-place major upgrades break inside containers, and how to sequence them safely.
Use Your Obsidian Vault from Neovim, Organized by Project
The workflow behind notes-nvim — project-scoped notes without leaving the editor.
Switch Your Samsung M8 Monitor HDMI Inputs from the Terminal (No Remote Needed)
If a workflow annoys me twice, I build something — even for monitor inputs.
cat stack.yml
cat contact.txt
$status --current
Open to remote roles · US-timezone overlap
$contact --email
email4abhisheksharma@gmail.com
$contact --links
linkedin.com/in/abhisheksharma91 · github.com/abhsss96
$