I build the backend that keeps AI products and money systems standing when the things they depend on fail.

SDE-2 at Eloelo, building Dolphin AI. Three years of Java, Spring Boot, Kafka and AWS, spent owning services end to end: design, code, deploy, on-call, and the 3am page.

Creator request image · video · audio Kafka SQS Router caps · retry fallback Provider 1 · image Provider 2 · image Provider 3 · 429 Provider 4 · video Provider 5 · video Provider 6 · audio Provider 7 · audio Provider 8 · upscale out
The Dolphin AI orchestration layer. Red is a provider rate-limiting us; the packet reroutes instead of failing the creator. 18% → under 5% end-to-end failure rate

Now

Building Dolphin AI at Eloelo

A multimodal GenAI platform that lets creators chain image, video and audio generation into one pipeline. I own the orchestration layer: async inference on Kafka and AWS SQS, routing across eight third-party model providers, and the observability that keeps it honest.

The interesting problems were never the API calls. They were what happens when a provider slows down, rate-limits us, or quietly returns something unusable. Per-provider concurrency caps, retries with budgets, fallback routing and Prometheus/Grafana dashboards took the end-to-end failure rate from 18% to under 5%. The platform is in internal beta under load and concurrency hardening ahead of public launch.

I also own prompt-quality iteration for the image and video models, so I see the whole loop from model output to what a creator actually gets.

14:02:11.408 INFO job=gen-7f3a kind=video provider=p4 accepted lane=2/6 14:02:12.955 WARN job=gen-7f3a provider=p4 status=429 retry_after=8s 14:02:12.957 INFO job=gen-7f3a fallback p4→p5 reason=rate_limited attempt=2/3 14:02:13.002 INFO job=gen-7f3a provider=p5 accepted lane=4/6 14:02:41.610 INFO job=gen-7f3a done latency=30.2s cost=0.041 route=p4!→p5 14:02:41.611 INFO metrics provider.p4.rate_limited+1 pipeline.fallback_success+1 14:02:44.120 ERROR job=gen-80c1 provider=p3 status=502 attempt=3/3 → dead_letter 14:02:44.121 INFO alert fired grafana:provider-p3-5xx > 2% (5m)

Systems I've shipped

Eloelo · 2026Dolphin AI

Multi-provider GenAI orchestration

Problem
Eight AI providers, each with its own latency, limits and failure modes, feeding one creator-facing product that couldn't afford to fail.
Built
Event-driven async inference on Kafka and SQS. Per-provider concurrency caps, bounded retries, fallback routing, dead-lettering, and Prometheus/Grafana telemetry on provider latency, queue lag and error rate.
Impact
End-to-end request failure rate down from 18% to under 5%.
Java · Spring Boot · Kafka · AWS SQS · Redis · Prometheus · Grafana
Eloelo · 2025Payments & Billing

Instant creator payouts with KYC and TDS

Problem
Creators waited on a manual finance process to get paid, and every payout had to be tax-correct and auditable.
Built
A payout service on the Cashfree gateway with automated KYC verification and TDS deduction, idempotent APIs, secrets management, and audit-ready records. Sole owner from design through on-call.
Impact
A fully manual finance workflow removed; payouts became instant and compliant by construction.
Java · Spring Boot · Cashfree · MySQL · AWS
Eloelo · 2025Engagement

Real-time leaderboard engine

Problem
Live rankings for thousands of creators, updated as scores streamed in, read constantly by the app.
Built
Redis sorted sets fed by a Kafka event stream, with write batching and read paths tuned for the hot ranges.
Impact
Sub-10ms p99 reads under 20K+ score updates per minute at peak.
Java · Kafka · Redis
Tekion · 2023–25Parts domain

Real-time CDC pipeline to search

Problem
Dealer search was drifting from the source of truth; about 5% of database changes never reached Elasticsearch.
Built
A custom Kafka connector for change data capture from MySQL and MongoDB into Elasticsearch, then rebuilt the consumers with partition-aware multi-threading.
Impact
2K+ events/sec across 50+ tables, event loss down to under 1%, consumer throughput up 10x, and Elasticsearch cost down 30% from query and index-merge work.
Java · Kafka Connect · MySQL · MongoDB · Elasticsearch

Experience

  1. May 2025 – present

    Software Development Engineer 2, Eloelo (StoryTV, Dolphin AI)

    Sole owner of four production microservices: Payments, Billing, Campaign Management and the Global Kafka Consumer, at 99.9% availability. Since June 2026, building Dolphin AI's orchestration layer.

  2. Jan 2023 – May 2025

    Tekion Corp, Bengaluru

    Intern to Associate Software Engineer to Software Engineer I. Built a production microservice from scratch in the first six months, then owned 5+ services in the Parts domain and the real-time data pipeline above.

  3. 2019 – 2023

    B.Tech, Information Technology, Manipal Institute of Technology

    CGPA 9.3/10. Somewhere in there, 800+ data-structures problems on LeetCode and GeeksforGeeks.

Stack

Languages and frameworks

  • Java
  • Spring Boot
  • Python
  • SQL
  • JavaScript
  • Bash

Data and messaging

  • Apache Kafka
  • Redis
  • PostgreSQL
  • MySQL
  • MongoDB
  • Elasticsearch
  • RabbitMQ
  • AWS SQS

Infrastructure and reliability

  • AWS
  • Docker
  • Kubernetes
  • Jenkins
  • GitHub Actions
  • Prometheus
  • Grafana
  • CloudWatch

How I work

Own it to the pager. A service isn't done when it's merged. It's done when I'd be comfortable being the one paged for it, and I usually am.

Design for the failure, not the demo. Every external dependency will rate-limit, time out, or lie eventually. The architecture decides whether that's a log line or an outage.

Measure before and after. 18% to 5%. 5% to 1%. 10x. If I can't put a number on the change, I'm not sure I made one.

Use AI, review everything. I use AI coding tools every day for boilerplate, tests and debugging. I don't let them make design decisions, and nothing they write reaches production unread.

Contact

Looking for backend, AI platform, or forward-deployed engineering roles at product companies building serious systems. Bengaluru or remote.

Download resume (PDF)
riteshranka.engi@gmail.com linkedin.com/in/ritesh-ranka github.com/ritesh-ranka