Backend & Systems EngineerNitte, Karnataka, India

Elvin RodriguesGo backends that hold up when the cache goes cold.

Two services in production on Go, PostgreSQL and Redis. The benchmarks, query plans and trade-offs are written down — including the known weaknesses. CS undergrad at NMAMIT, DSA Advisor at Finite Loop Club.

NowConcurrent Distributed Job Queue — Go · FOR UPDATE SKIP LOCKED

GET /:codetrimto.me

The short code is already in Redis.

Ingressnet/http · MaxBytesReader1 req
Rate limitRedis Lua · sliding-window ZSET1 round-trip
CacheRedis cache-aside, allkeys-lruHIT
Singleflightin-process request coalescingskipped
PostgreSQLcovering index · index-only scanskipped
Response302 Found → destination302

Concurrent in1

DB queries0

Redirectsub-5ms

Path and figures from the URL shortener's committed benchmarks. How it was measured

Measured, not estimatedhey · go test -race · EXPLAIN ANALYZE

Cache stampede: 1DB query per 1,000Simultaneous misses on one cold key collapse into a single read.
Sustained load: 195,357redirects, 0 errors60-second soak at 200 workers — 3,253 req/s, p99 139ms.
Singleflight A/B: +35%throughput5,871 vs 4,351 req/s on the same build, coalescing off then on.
Deploy artefact: 12 MBscratch containerCGO-free static binary, down from a 604 MB build image.

01 / About

I care about what a service does on its worst day — when the cache is cold, the database is contended, and a dependency has stopped answering.

Most of what I know came from building two services end to end and then trying to break them: killing Redis in the middle of a benchmark run, firing a thousand concurrent requests at one expired key, reading the query plan instead of guessing at it. The interesting part was never the happy path.

Each project ships with a written case study — the problem, the trade-off I chose and why, the schema, and the benchmark output. Where something has a known weakness the write-up says so: the shortener's click counter still dirties the heap page that its covering index was built to avoid, and the fix is documented rather than hidden.

Alongside that I run DSA training for 150+ students at Finite Loop Club, which is where most of the C++ and the habit of explaining a trade-off out loud comes from.

Dossier

Core focus: Concurrency, queues, database internals
Languages: Go for backend · C++ for DSA
Problem solving: 600+ solved · ~1700 contest rating
Campus leadership: DSA Advisor, Finite Loop Club
Academics: NMAMIT, Nitte (CSE '27) · 8.79 CGPA
Availability: Internships now · Full-time 2027

02 / Stack

A narrow toolchain, used properly.

Not a logo wall. Each entry below is something I have reached for on the hot path of a service that is running right now, and can explain the failure mode of.

Runtime & Concurrency

Go runtime mechanics & synchronization

Go (Golang)Primary backend language
Goroutines & ChannelsCSP concurrency model
Singleflight CoalescingThundering herd suppression
Sync PrimitivesMutex, RWMutex, WaitGroup, Once
Context DisciplineCancellation & timeout trees
Race Detectiongo test -race verification

Databases & Storage

Query optimization & transaction guarantees

PostgreSQL 15Schema design & EXPLAIN ANALYZE
Covering IndexesIndex-only scan optimization
Partial Unique IndexesSoft-deletion uniqueness
Redis CachingCache-aside & stampede guard
Redis Lua ScriptsAtomic sliding-window counters
Parameterized SQLNo ORM, injection-safe

Reliability & Tooling

Production resilience & deployment

Linux / Arch CLIProcess management & tooling
Docker Scratch12MB CGO-free static binaries
GitHub ActionsAutomated test & build pipelines
Fail-Open DegradationCache outage survival
Health ProbesDecoupled /healthz and /readyz
Load Testinghey & vegeta benchmarking

Foundations & CS

Algorithmic intuition & distributed design

C++600+ solved · ~1700 contest rating
System DesignLayered architecture & trade-offs
Operating SystemsConcurrency, scheduling & IO
Computer NetworksHTTP, TCP/IP, connection pools
Timing-Safe AuthConstant-time bcrypt verification
REST API ContractsChi Router v5 JSON envelopes

03 / Systems

Three systems, and what each one costs.

Every entry links to a written case study: the problem, the trade-off and its rationale, the schema, and the raw benchmark output. Where something is unfinished or has a known weakness, the write-up says so.

Concurrency & CachingLivetrimto.me

High-Throughput URL Shortener

URL shortening engine live at trimto.me — Go + PostgreSQL + Redis, with singleflight cache-aside, Lua rate limiting, and the benchmarks committed.

How it works

Singleflight Request CoalescingA thousand simultaneous requests for the same expired link become one database query instead of a thousand. Measured at +35% throughput versus coalescing off.
Fail-Open Under Cache OutageIf Redis disappears the redirect path falls back to PostgreSQL rather than returning 500s. Benchmarked with Redis killed mid-run.
Covering Index for Heap-Free ReadsThe lookup index carries the destination in its leaf pages, so a redirect resolves without touching the table — with a documented caveat about the click counter.

Measured

Singleflight A/B: +35% throughput5,871 vs 4,351 req/s; p99 448ms → 364ms (controlled, same build)
Stampede Guard: 1 DB queryPer 1,000 concurrent misses on the same cold key
60s Soak: 195,357 reqs3,253 req/s sustained, p99 139ms, zero errors
Binary Footprint: 12 MBCGO-free static scratch container

GoPostgreSQLRedisDockerRate LimitingCachingSingleflight

Read the case study Live system Source

Clean Architecture & SecurityLivecontact-manager-fawn-alpha.vercel.app

ContactHub Backend Service

Full-stack contact management engine live at contact-manager-fawn-alpha.vercel.app — Go clean architecture, bcrypt crypto, and partial unique indexing.

How it works

Timing-Safe Account Enumeration DefenceAn unknown email still pays the cost of a bcrypt comparison against a dummy hash, so response time cannot be used to discover which accounts exist.
Partial Unique Index on Soft DeletePhone uniqueness is scoped per user and to live rows only, so a deleted contact can be recreated or restored without colliding with its own tombstone.
Instant Session Revocation Without RedisA token_version column in PostgreSQL is carried in the JWT claims; a password reset increments it and every previously issued token stops validating.

Measured

Clean Architecture: 4 LayersHandler → Service → Repo → Domain
Auth Defense: Timing-SafeConstant-time dummy bcrypt verification
Data Isolation: Partial IndexScoped WHERE deleted_at IS NULL
Shipped Endpoints: 20 APIsFull-stack CRUD, Auth, Admin & Health

GoPostgreSQLDockerJWTREST APIClean Architecturebcrypt

Read the case study Live system Source

Distributed Systems & ConcurrencyIn Progress

Concurrent Distributed Job Queue

Go + PostgreSQL queue engine using FOR UPDATE SKIP LOCKED for contention-free job claiming, with priority dispatch and orphan recovery.

How it works

Contention-Free Claiming with SKIP LOCKEDWorkers skip rows another worker already holds instead of queueing behind them, so adding workers adds throughput. A test asserts exactly one worker claims a given job.
Crash Recovery via Orphan ReaperJobs left in processing past a timeout are swept back to pending. Simpler than a heartbeat protocol, and it fails safe: a job may run twice, so processors must be idempotent.
Per-Job Panic RecoveryA panicking processor is recovered and recorded as a job failure rather than taking down the worker goroutine and the rest of the pool with it.

Measured

Worker Contention: SKIP LOCKEDWorkers never block on each other's rows
Execution Engine: Goroutine PoolContext cancellation & graceful shutdown
Crash Recovery: Orphan ReaperTimed-out claims return to pending

GoPostgreSQLDistributed SystemsConcurrencyQueues

Read the case studySource on request

04 / Milestones

Where the practice came from.

Campus technical leadership at Finite Loop Club and a computer science degree at NMAMIT, in reverse order.

2026 — PresentAdvisory Board

DSA Advisor

Finite Loop Club · NMAMIT

Elevated to the senior Advisory Board after serving as DSA Coordinator. Directing algorithmic training, structuring advanced dynamic programming and graph curriculums, and mentoring 150+ student engineers in competitive problem-solving.

2025 — 2026Core Committee

Core Committee · DSA Coordinator

Finite Loop Club · NMAMIT

Selected into the student Core Committee after a year of active club contributions. Coordinated weekly campus algorithmic workshops, organized coding contests, and delivered sessions on time-space complexity and foundational data structures.

2024 — 2025Member

Club Member

Finite Loop Club · NMAMIT

Joined Finite Loop Club as an active member, participating in campus hackathons, weekly competitive coding rounds, and collaborative peer learning circles.

2023 — 20278.79 CGPA

B.Tech, Computer Science & Engineering

NMAM Institute of Technology, Nitte

Coursework in Operating Systems, DBMS, Computer Networks and Systems Software, alongside 600+ solved algorithmic challenges and a ~1700 LeetCode contest rating.

05 / ContactOpen

Hiring for a backend team? Let's talk.

I am looking for a backend engineering internship now, and full-time work from 2027. Based in Nitte, Karnataka — open to relocating or working remotely. The fastest way to reach me is email; I reply to everything.

elvinrodrigues3456@gmail.com

Elsewhere