Backend-focused engineer with 8+ years designing and scaling multi-tenant SaaS platforms across petroleum logistics, eCommerce, and non-profit sectors. I build systems end-to-end — from plugin frameworks and RAG-based AI pipelines to billing integrations and real-time dispatch — and I care about the unglamorous parts too: query plans, infra bills, test coverage, and team standards.
$5,000 → $1,200/month at FleetPanda via API, query, and call-pattern optimisation
Eliminated N+1 bottlenecks on the most-trafficked application views
pg-vector chatbot pipeline — embeddings, retrieval, scheduled refresh — shipped as a pro-tier feature
Platforms serving hundreds of businesses across logistics and eCommerce
A few systems I designed or owned, and the problem each one solved
Problem: Every customer wanted their own back-office tool (QuickBooks and others) wired into the platform, and each bespoke integration added coupling to the core app.
Approach: Designed an open-source, installable plugin system. Each droplet is an autonomous service with its own database that owns its logs, callbacks, and field-mapping logic, behind a published spec.
Outcome: Integrations moved out of the monolith, and third-party vendors can now build their own droplets against the spec.
Problem: Customers needed an AI assistant grounded in their own constantly changing catalogue and content, not generic LLM answers.
Approach: Built the full pipeline: pg-vector embedding storage, retrieval logic, and an automated refresh strategy using debounced Sidekiq jobs so bursts of edits don't trigger redundant re-embedding.
Outcome: Shipped to production as a pro-tier, revenue-bearing feature.
Problem: Heroku costs had grown to $5,000/month while the most-used views were getting slower.
Approach: Profiled with ScoutAPM, New Relic, and Coralogix; removed N+1 queries, tuned slow queries and APIs, added database views and pagination, and gave the frontend team structured feedback on excessive call patterns.
Outcome: Server costs fell to $1,200/month and database performance improved by 30%.
Problem: Fuel distributors ran dispatch, invoicing, and driver management across phones, spreadsheets, and disconnected tools.
Approach: Owned the multi-tenant backend: REST and GraphQL APIs with RBAC, Twilio calling and SMS to reach drivers in the field, a HelloSign-powered hiring ATS, and integrations with DTN, QuickBooks, and tank monitors.
Outcome: Served 20+ fuel distributors and hundreds of active fleets on one platform.
Problem: A Rails 4 schema-per-tenant design made every migration, deploy, and cross-tenant report slower as the customer count grew.
Approach: Planned and executed a move to Rails 6 with single-schema, row-scoped tenancy, with strict data segregation.
Outcome: Simpler operations and a codebase that could keep up with the customer count.
Problem: Receipts arriving by email had to be read and entered by hand.
Approach: Designed a serverless pipeline (SES → S3 → EventBridge/SQS → Lambda → RDS) with a Python parser and a Rails portal for reviewing and correcting results.
Outcome: Inbound receipts are processed automatically at scale, and people only step in when a result needs correcting.
8+ years building and scaling production SaaS for US-based companies, remotely
Part-time, alongside deliberate upskilling in AI agents and LLM tooling
Senior engineer across multiple product pods in a distributed team, delivering backend features end-to-end
Joined as an early engineer; grew into solutions architect over 3+ years
Tools and patterns I use in production
Academic background and professional certifications
Computer Engineering
Kathmandu University, Nepal
GPA: 3.64 out of 4.0September 2016
+2 Science
Radiant Higher Secondary School, Nepal
May 2012