dhanaraj = {
    "title"       : "Result-Driven · Event-Driven Software Engineer · Published NLP Researcher",
    "identity"    : [
        "I don't write code — I engineer outcomes.",
        "Every system I build is async, reactive, and built to survive at scale.",
        "Events flow. Services decouple. Results ship.",
        "I research what I build — and build what I research.",
    ],
    "published_at": "Dublin City University — MSc Computing, Secure Software Engineering",
    "research"    : "NLP · Semantic Clustering · LLMs · RAG · Influencer AI Systems",
    "architecture": "Event-Driven · Distributed · Microservices · CQRS · Saga · Kafka",
    "stack"       : "Java · Python · React · React Native · Node.js · Go",
    "devops"      : "Kubernetes · Docker · Terraform · CI/CD · GitOps · Prometheus",
    "targeting"   : ["Meta", "Amazon", "Apple", "Netflix", "Google"],
    "philosophy"  : "Ship for impact. Design for failure. Scale without mercy.",
}

Active on: LeetCode · TopCoder · Codeforces · HackerRank

Strongest areas: Dynamic programming · Graph algorithms · Segment trees · Binary search · Hard problems · System design

Five production-grade systems. Each one event-driven. Each one solving a real problem. Each one built the way MAANG engineers build.

Problem → Unprotected APIs get hammered. Abuse, DDoS, runaway clients destroy infra.
Solution → Distributed, Redis-backed rate limiter. Decision made in < 1ms. Every time.

Results:

• ✦ 100k+ req/sec throughput with sub-millisecond decision latency
• ✦ Zero downtime under Redis node failure — automatic cluster failover
• ✦ Pluggable algorithm engine: token bucket · sliding window · fixed window

Event-driven core: Rate limit events → Kafka → analytics consumer → Grafana real-time abuse dashboard

Stack: Java Spring Boot Redis Cluster Kafka Docker Kubernetes Prometheus Grafana

Problem → Algorithms are invisible. Reading quicksort a hundred times ≠ understanding it.
Solution → Make every comparison, swap, and partition visible in real time.

Results:

• ✦ 6+ algorithms with step-through mode and live Big O complexity panel
• ✦ Swap and comparison counters — quantify algorithm cost in real time
• ✦ Actively used as a DSA prep and teaching tool

Stack: React TypeScript Web Animations API Vite

Problem → Applying to 50 jobs takes 50 hours of copy-paste. No engineer should do that.
Solution → AI agent that finds, tailors, and applies — while you sleep.

Results:

• ✦ Fully autonomous end-to-end — zero manual intervention per application
• ✦ LLM + RAG pipeline tailors resume per job description
• ✦ Runs 24/7 unattended on Kubernetes

Event-driven core: Scraper event → Kafka → relevance scorer → LLM tailoring worker → submission service → status event → React dashboard

Stack: Python LangChain OpenAI API RAG Playwright Selenium React Kafka Kubernetes

Problem → Refreshing job boards 20 times a day. Missing the perfect role hours after it posts.
Solution → Push-based alert system. The job board comes to you — instantly, on your favourite app.

Results:

• ✦ < 30 second latency from job posting to push notification delivery
• ✦ Multi-platform: Telegram · Slack · WhatsApp · Discord
• ✦ Intelligent deduplication — zero spam, only relevant alerts

Event-driven core: job.posted → Kafka → filter service → notification dispatcher → platform webhook → delivery receipt → DLQ retry on failure

Stack: Python Kafka Redis FastAPI Telegram Bot API Slack API Webhooks Docker Kubernetes

Problem → Influencers drown in thousands of DMs. Followers get ghosted. Engagement collapses.
Solution → Research-backed AI platform. Reads every message. Responds like the influencer.

This project is the production implementation of our peer-reviewed published paper — grounded in academic research and real-world validation.

Results from research evaluation:

• ✦ 63.93% clustering accuracy on Quora Question Pairs — zero supervision, zero labelled data
• ✦ Sentence-BERT outperformed TF-IDF baseline on semantic paraphrase detection
• ✦ HDBSCAN dynamically adapts — no predefined number of clusters required

Event-driven core: message.received → Sentence-BERT encoder → HDBSCAN cluster → RAG retrieval → LLM reply → confidence score → auto-send or human-review queue → reply.sent → analytics pipeline

Stack: Python Sentence-BERT HDBSCAN LangChain RAG OpenAI API FastAPI React PostgreSQL Redis Kafka Docker

Peer-reviewed, published, and cited — engineering backed by academic rigour.

Transforming Influencer-Follower Engagement with AI-Powered Response Systems Using Natural Language Processing Dhanaraj Sujatha · Dhanushkumar S G Dublin City University — MSc Computing, Secure Software Engineering · July 2025 Abstract: Social media influencers receive hundreds of repetitive follower questions daily — manually responding is unsustainable. This paper presents a semi-automated Q&A pipeline that solves the scalability problem while preserving authentic, personalised engagement at scale. Three-stage modular architecture: Stage 1 — Semantic Question Clustering: Questions encoded via Sentence-BERT (all-MiniLM-L6-v2) and grouped with HDBSCAN — achieving 63.93% clustering accuracy across 10 test sets on Quora Question Pairs, with zero supervision or labelled training data. Stage 2 — Merged Question Generation: OpenAI GPT-3.5-turbo with few-shot prompting consolidates each cluster into one coherent representative question, reducing influencer cognitive load without losing follower intent. Stage 3 — Personalised Reply Generation via RAG: The influencer's single answer is stored in a persistent knowledge base. Future similar questions trigger retrieval and LLM-crafted on-brand personalised replies. Key contributions: Modular NLP pipeline · Sentence-BERT embeddings · HDBSCAN density clustering · RAG knowledge retrieval · LLM prompt engineering · Human-in-the-loop design Stack: Python Sentence-BERT HDBSCAN OpenAI API RAG Django React

• InfluenceIQ → extending the published research into multi-agent orchestration with MCP-based persistent memory
• JobPilot + JobAlert → merging into one unified autonomous job-hunting platform
• Mastering distributed event systems — Raft, Paxos, vector clocks, CRDTs, exactly-once semantics
• Designing for MAANG scale — 100M users, petabyte event logs, five-nines availability
• Daily LeetCode Hard + contest problems — consistent, relentless, every day

Result-Driven. Event-Driven. Research-Backed. Built for Scale.

Open to senior engineering roles at Meta · Amazon · Apple · Netflix · Google

Let's build systems that react, scale, and actually matter.