╔══════════════════════════════════════════════════════════════╗
║          ASHISH BABU Z  ·  BACKEND ENGINEER                  ║
║          Distributed Systems · RAG · AI Integration          ║
╚══════════════════════════════════════════════════════════════╝

data class Engineer(
    val name: String       = "Ashish Babu Z",
    val role: String       = "Backend Engineer",
    val location: String   = "Chennai, India 🇮🇳",
    val status: String     = "Immediate Joiner · Open to Remote",
    val experience: String = "Nearly 3 years · Series C SaaS · 1,000+ Enterprise Clients",
    val superpower: String = "Zero post-release bugs on security-critical, money-critical systems"
)

Things I've shipped that actually mattered

a4f9c2e  [BILLING]   Owned billing/overage engine → 500-1,000+ MSP clients platform-wide
8b3d1f7  [SSO]       Solo multi-tenant SAML 2.0 SSO → 1,000+ clients, zero post-release bugs
c91e3a2  [PERF]      Eliminated N+1 patterns via APM → 90%+ latency reduction
2f7b8d4  [BUG]       Traced customer-reported billing bug through 20mo of prod logs → root-caused a race condition, corrected all 12 affected assets
e5a1c9f  [ASYNC]     Fixed silent timezone bug via payload metadata → zero missed alerts
7d4e2b1  [PRODUCT]   3-strike dismissal logic on ISL prompt → shipped as designed

Self-implemented Raft consensus engine & distributed lock manager

Leader Election → Log Replication → Majority Commit
     → Distributed Lock Manager (HTTP acquire/release/renew)
          → Verified: 50-client concurrent load, exactly 1 success

Stack: Kotlin gRPC Docker

• 🗳️ Leader election with randomized timeouts, majority voting
• 📝 Log replication with majority-based commit
• 🔒 Conflict-safe distributed lock manager — lease expiry, disk persistence, leader-only writes
• ✅ Verified live: automatic failover on leader kill, 50-client concurrent load test (1 success, 49 correctly rejected)

⌥ GitHub · ↗ Read the write-up

RAG-powered semantic job matching platform

PDF Resume → HuggingFace Embeddings → pgvector Storage
     → Cosine Similarity Search → Color-coded Match Scores
          → LLM Gap Analysis via OpenRouter

Stack: Kotlin Spring Boot 3.3.5 PostgreSQL pgvector HuggingFace OpenRouter Docker

• 🧠 End-to-end RAG pipeline — PDF parsing, vector embeddings, semantic search
• 🎯 GREEN/ORANGE/RED match scoring against 60+ real job listings
• 🤖 AI gap analysis — what you have, what you're missing, in plain English
• ⚙️ Dual provider — Ollama locally, HuggingFace + OpenRouter in production
• 🐛 Found and fixed a scoring-integrity bug — the LLM independently guessed a match % that could disagree with the already-computed cosine similarity score; fixed by injecting the real score into the prompt as ground truth. Later, a reader flagged a second risk (same-dimension embeddings from different models silently producing meaningless comparisons) — added model identity tracking so mismatched or missing model IDs are now rejected outright.

⌥ GitHub · ↗ Read the write-up

AI-powered cinema booking platform · Hexagonal Architecture

Mood Input → Llama 3.3 70B → Ranked Recommendations
     → Redis Seat Lock (TTL) → Idempotent Booking
          → Rev-Coins Reward → Confirmed Ticket

Stack: Kotlin Spring Boot 3.2 PostgreSQL Redis WebSocket/STOMP Llama 3.3 70B Resilience4j

• 🏗️ Hexagonal architecture — domain layer has zero Spring/JPA dependencies
• 🔒 Atomic Redis seat locking — 10-min TTL, NX command, race condition found and fixed under real concurrent load
• 🤖 Rev-Bot AI concierge — mood-based recommendations via WebSocket
• 💰 Rev-Coins reward engine — earn on booking, review, ad watch
• ⚡ Resilience4j circuit breaker — fallback ensures zero downtime

⌥ GitHub · ↗ Read the write-up

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

Token Bucket → Goroutine-safe Map Access (mutex)
     → Idle-client Eviction → Context Cancellation

Stack: Go Docker

• 🪣 Per-client token bucket rate limiter
• 🔐 Goroutine-safe map access via mutex
• 🧹 Idle-client eviction via background goroutine + context-based cancellation
• ✅ Verified with a custom test: 75 concurrent requests across 5 clients, zero data corruption

⌥ GitHub

Technical write-ups on real bugs and system design decisions — not tutorials, post-mortems

• 📝 I Broke My Own Cluster With a Node That Could Never Win — a real production-style failure mode in Quorum's leader election, found and fixed
• 📝 Locking a Seat Without Losing the Sale — a TOCTOU race hiding behind an atomic Redis primitive, reproduced with a concurrency test and fixed
• 📝 I Let My LLM Guess a Number My Own Code Already Knew — an LLM and a cosine-similarity score disagreed on the same match, and why only one of them should ever get to answer

languages:
  primary:   [ Kotlin, Java, Go, SQL ]

frameworks:
  backend:   [ Spring Boot, Spring Security, Spring WebFlux ]
  orm:       [ JPA, JOOQ, Hibernate ]
  api:       [ RESTful, GraphQL, WebSocket/STOMP ]

databases:
  relational: [ MySQL, PostgreSQL, pgvector ]
  cache:      [ Redis ]

security:
  protocols: [ SAML 2.0, OAuth 2.0, JWT, TOTP ]
  encryption: [ AES-256 ]
  patterns:  [ RBAC, Audit Logging, Idempotency ]

messaging:
  brokers:   [ Apache Pulsar, Kafka ]
  patterns:  [ Event-Driven, Avro, Async Processing ]

ai_and_llm:
  patterns:  [ RAG, Vector Embeddings, Cosine Similarity ]
  providers: [ OpenRouter, HuggingFace, Ollama ]
  models:    [ Llama 3.3 70B ]
  tools:     [ pgvector, Apache PDFBox, Prompt Engineering ]

observability:
  tools:     [ Coralogix, APM Profiling, Slack Alerting ]
  patterns:  [ Spring Events, Circuit Breaker, Health Checks ]

devops:
  cloud:     [ AWS S3, Lambda, RDS ]
  containers: [ Docker, Docker Compose ]
  ci_cd:     [ Jenkins, Git, Gradle ]
  migrations: [ Liquibase ]

testing:
  frameworks: [ JUnit, Mockito, MockK ]
  approach:   [ TDD, concurrency testing ]

architecture:
  patterns:  [ Hexagonal, Ports & Adapters, Resilience4j ]
  principles: [ Zero post-release bugs, Idempotency-first, Domain isolation ]

$ echo $ENGINEERING_PHILOSOPHY

→ Money can never be wrong.
→ If it's not idempotent, it's not done.
→ Own it end-to-end or don't touch it.
→ Zero post-release bugs is a choice, not luck.
→ The domain should never know the infrastructure exists.
→ Trust, but verify — test the thing, don't just reason about it.

Chennai, India · Immediate Joiner · Open to Remote
ashish.babu.sde@gmail.com · linkedin.com/in/ashish-babu-z

Building things that work. Every time.