Ashish Babu

Ashish Babu

Building scalable systems at the intersection of backend engineering and AI

Nearly 3 years building production-grade distributed systems at SuperOps.ai (Series C, $54M) — serving 1,000+ enterprise clients globally. Self-implemented a Raft consensus engine from scratch, shipped multi-tenant SSO with zero post-release bugs, and built RAG pipelines integrating LLMs into production and personal systems. Based in Chennai. Immediate joiner.

✓ Immediate Joiner ✓ Open to Bengaluru / Remote Kotlin · Java · Spring Boot PostgreSQL · Redis · Kafka RAG · pgvector · LLMs

Problems I've Solved

Race Condition in Billing Pipeline

A customer flagged a billing discrepancy. I independently traced the defect through 20 months of Coralogix production logs and found a systemic race condition between asset hard-deletion and the billing/usage update logic — deleted assets kept getting charged. Identified every affected asset platform-wide and fixed the underlying race.

→ All 12 affected assets identified & corrected · Root cause fixed platform-wide

SAML 2.0 SSO — Solo Delivery at Scale

Most-requested enterprise feature. I was the sole engineer — designed per-client IDP configuration, RBAC enforcement, and a dual-write migration strategy to support existing tenants without downtime. Shipped across Mobile App, Requester Portal, and Admin Portal.

→ 1,000+ tenants · Zero post-release bugs · Multi-phase rollout

Policy Duplication Race Bug

A frontend double-fire on createPolicy was bypassing single-policy-per-type validation, silently creating duplicate policies and blocking all future overrides for affected clients. Fixed the race condition and designed a Strategy-pattern validation interface across six policy categories.

→ Eliminated recurring class of bad-data bugs platform-wide

N+1 Cache Call Killing API Performance

Policy creation endpoints were making a Redis feature-gate call on every iteration — N+1 pattern. Found it via APM profiling on Coralogix, refactored to a single cached object per request lifecycle, and built dashboards + Slack alerting for ongoing production monitoring.

→ 90%+ API latency reduction

Silent Timezone Bug in Async Flows

Background threads in the bulk action engine were losing timezone context during async execution — operations running at wrong times for clients in different zones. Fixed by embedding timezone metadata directly into the message payload before dispatch.

→ 20,000+ records/run · 90% operation time reduction

MDM Silent Monitoring Failure

While writing test cases for MDM policy management, I caught a silent defect where a batch of client assets were going completely undetected and unmonitored — a churn risk no one had noticed. Fixed it before any customer was impacted.

→ Caught before customer impact · Prevented potential churn

What I've Built

Quorum

Self-implemented Raft consensus engine & distributed lock manager

● VERIFIED LIVE

Built the core of a distributed consensus system from first principles — not a wrapper around an existing library. Leader election, log replication, and a conflict-safe distributed lock manager on top, verified under real node failure and concurrent load.

Leader election with randomized timeouts and majority voting
Log replication with majority-based commit
Dynamic cluster membership with overlap-safety guard
HTTP acquire/release/renew lock API — lease expiry, disk persistence
50-client concurrent load test: exactly 1 success, 49 correctly rejected
Automatic failover verified live on leader kill

Kotlin gRPC Docker Raft Consensus Distributed Locking

⌥ GitHub ↗ Read the write-up

ProximaHire

RAG-powered semantic job matching

● LIVE

Eliminates the ATS keyword problem by matching resumes to jobs using vector embeddings and cosine similarity — not keywords. Built a full RAG pipeline from PDF parsing to LLM gap analysis.

PDF resume parsing via Apache PDFBox
384-dim vector embeddings via HuggingFace (all-MiniLM-L6-v2)
Cosine similarity search using pgvector on 60+ job listings
LLM gap analysis via OpenRouter — identifies missing skills
Dual provider — Ollama locally, HuggingFace + OpenRouter in prod
Found and fixed two real bugs: an LLM independently guessing a match % that disagreed with the computed cosine similarity score, and a silent cross-model embedding comparison risk flagged by a reader — added model identity tracking to guard against it

Kotlin Spring Boot 3.3.5 PostgreSQL pgvector HuggingFace OpenRouter Docker RAG

↗ Live Demo ⌥ GitHub ↗ Read the write-up

Reverix 2.0

AI-powered mood-based movie booking

● LIVE

Full-stack movie booking platform with AI chatbot, atomic Redis seat locking, and hexagonal architecture. Rebuilt from scratch after identifying structural limitations in v1.

Hexagonal architecture — pure domain layer, zero framework deps
Atomic Redis seat locking via NX command, 10-min TTL
Idempotency key guard on every booking transaction
RevBot AI chatbot — Llama 3.3 70B via WebSocket/STOMP
Rev-Coins reward engine — earn-and-redeem loop across bookings, reviews, ads
Resilience4j circuit breaker + self-ping cron for 24/7 free-tier uptime
Found and fixed a check-then-act race condition in multi-seat locking, reproduced with a 20-thread concurrency test and verified fixed (1 success, 19 correctly rejected)

Kotlin Spring Boot 3.2 Redis PostgreSQL Llama 3.3 70B WebSocket Resilience4j Liquibase

↗ Live Demo ⌥ GitHub ↗ Read the write-up

Ratchet

Concurrent rate limiter, built to prove hands-on Go proficiency

● VERIFIED

Per-client token bucket rate limiter, written in Go specifically to validate real Go proficiency beyond a resume line — goroutine-safe under concurrent load, verified with a custom concurrency test.

Goroutine-safe map access via mutex
Idle-client eviction via background goroutine + context cancellation
Verified with 75 concurrent requests across 5 simultaneous clients — zero data corruption
Containerized with Docker

Go Concurrency Docker

⌥ GitHub

Experience

Software Engineer — Backend Engineering

SuperOps.ai · Series C · $54M Funded · 1,000+ MSP Clients

Full Time

Jun 2024 – Mar 2026
Chennai, India

Race Condition Fix (Revenue-Critical): Investigated a customer-reported billing discrepancy and independently traced it through 20 months of Coralogix production logs to a systemic race condition between asset hard-deletion and billing/usage update logic — identified all 12 affected assets platform-wide, fixed the underlying defect, and corrected charges.
Billing Calculation Engine: Owned billable and overage asset calculation logic for 500–1,000+ clients — engineered grace-period and month-2-onward billing rules by activation window. Automated monthly AWS S3 archival of calculation results.
SAML 2.0 SSO (Solo): Sole engineer for multi-tenant SSO across Mobile App, Requester Portal, and Admin Portal for 1,000+ tenants — per-client IDP config, RBAC, dual-write migration strategy. Zero post-release bugs.
Performance Optimization: Cut API response time 90%+ — eliminated N+1 feature gate cache calls via APM profiling, refactored to single cached object per request lifecycle. Built Coralogix dashboards and Slack alerting.
Bulk Action Engine: Filter-aware batch engine supporting 20,000+ records/run — async queue-backed for large selections, synchronous for small. 90% operation time reduction. Fixed silent timezone bug in async flows.
Policy Engine: Fixed duplicate-createPolicy race condition silently blocking client overrides. Designed Strategy-pattern validation interface across six policy categories.
ISL Online (Product Initiative): Proposed and shipped technician-level dismissal tracking with 3-strike suppression logic to cut onboarding-prompt friction. Built one-click ISL enablement — auto-provisioning a dummy subscription and required policy on activation, removing manual setup for customers.
Internal AI Tooling: Contributed to a Claude-powered multi-agent system generating and self-reviewing JUnit/Mockito test cases directly from the codebase across four collaborating agent roles (SDE, SDET, Manager, Testing Manager). Participated in early testing of an internal MCP server for the RMM module.

Software Engineering Intern — Backend

SuperOps.ai · Promoted to full-time in 12 months

Intern → FTE

Jun 2023 – Jun 2024
Chennai, India

Built internal bulk-update tool for 3,000+ assets/client — reduced multi-day manual onboarding to minutes.
Introduced Redis caching on high-frequency agent endpoints — improved response consistency under concurrent traffic across 1,000+ clients.
Built async event pipeline via Apache Pulsar streaming user events to Mixpanel for real-time product analytics.
Extended patch system from Windows/macOS to Linux across a 375,000+ patch platform.

Skills

Languages

KotlinJavaGoSQL

Frameworks

Spring BootSpring SecuritySpring WebFluxGraphQLJPAJOOQHibernate

Databases

PostgreSQLMySQLRedispgvector

Distributed Systems

Raft ConsensusLeader ElectionLog ReplicationCAP Trade-offsConcurrency Control

AI & LLM

RAG PipelinesVector EmbeddingsHuggingFaceOpenRouterLlama 3.3 70BMulti-Agent SystemsMCPPrompt Engineering

Messaging & Async

Apache PulsarKafkaAvroWebSocket/STOMPEvent-Driven

Cloud & DevOps

AWS S3AWS LambdaAWS RDSDockerKubernetesJenkinsLiquibase

Auth & Security

SAML 2.0OAuth 2.0JWTRBACTOTP2FARate Limiting

Architecture

HexagonalMicroservicesDistributed LockingResilience4jIdempotencyStrategy Pattern

Observability

CoralogixAPM ProfilingSlack AlertingSpring Events

Testing

JUnitMockitoMockKTDD

Impact at Scale

Problems I've Solved

What I've Built

Experience

Skills

Writing

Let's Connect