Knock knock.

Hello
Entering gateway automatically...
AVAILABLE FOR ROLES & AI COLLABORATIONS · COIMBATORE, TN [IN]

Hi, I'm Niranjan Murugarasu

AI/ML Systems Engineer & Product Builder

I'm Niranjan Murugarasu — an AI/ML Systems Engineer & Product Builder bridging deep generative AI capabilities with production-grade backend architecture. From fine-tuned clinical LLMs with deterministic safety guardrails to high-throughput Spring Boot backends and automated triage engines, I build scalable systems that deliver measurable business ROI.

500+
LeetCode Solved
5
Shipped AI Projects
8.65
B.Tech CGPA
2027
Graduation Year
→ See My Work Get In Touch
Niranjan Murugarasu — AI/ML Builder
⚡ NIRANJAN MURUGARASU
TECHNICAL ARCHITECTURE

Tech Stack & Core Arsenal

Production-proven frameworks, fine-tuned LLM tooling, distributed backend engines, and core computer science fundamentals.

GenAI & LLMs
Prompt Engineering LLM Evaluation & Validation RAG Architecture Vector Databases LoRA / QLoRA Fine-Tuning Hugging Face Hugging Face Transformers
ML & Data Science
PyTorch PyTorch Scikit-learn Scikit-learn Supervised & Unsupervised Learning NLP Pipelines Feature Engineering
Cloud & MLOps
GCP Google Cloud (GCP) Docker Docker Containerization Kubernetes Kubernetes Automated CI/CD Model Serving & Deployment
Programming & Databases
Python Python 3.12+ Java Java 25 / Spring Boot 3 SQL PostgreSQL 17 & SQL
Frameworks & Tools
FastAPI FastAPI Streamlit Streamlit Git/GitHub Git & GitHub
Core CS Fundamentals
Data Structures & Algorithms Object-Oriented Programming (OOP) Database Management Systems (DBMS) Operating Systems Distributed System Design
Applied Mathematics
Probability & Random Processes Statistical Inference & Statistics Linear Algebra & Matrices Convex & Mathematical Optimization
FEATURED PROJECTS

Shipped Work & Architecture

FIG. 01 SPEC: CLINICAL DRUG RECOMMENDATION LLM · 2026

MedRx Phi-4-mini · QLoRA · Hugging Face

Fine-Tuned Clinical Drug & Treatment Recommendation LLM
  • Built a post-inference output-validation layer running automated drug-allergen cross-reactivity and dosage range checks against a Pydantic-parsed schema to catch unsafe model outputs before reaching end users.
  • Designed a 7-gate data-quality filtering pipeline (PHI scrubbing, MD5 deduplication, drug-class-templated synthetic augmentation) reducing a 172K-record raw corpus to 13.7K high-quality training samples (92% rejection rate).
  • Fine-tuned Microsoft Phi-4-mini-Instruct (3.8B) via 4-bit QLoRA (NF4) with Unsloth, training an 8.9M-parameter LoRA adapter (0.23% of base) on the resulting 13,728-sample clinical-reasoning dataset.
  • Root-caused and resolved a Triton JIT compiler segfault specific to Ada Lovelace GPUs via native stack-trace analysis; published model publicly on Hugging Face.
Hugging Face Model ↗
class ClinicalSafetyGuard:
def verify_prescription(self, rx_data):
# Pydantic schema: dosage & allergen check
valid = self.schema_validator.check(rx_data)
if not valid:
return SafetyResult(status="REJECTED")
return self.llm.recommend(rx_data)
FIG. 02 SPEC: TICKET TRIAGE & ROUTING ENGINE · 2026

SwiftDesk AI Spring Boot 3 · React 19 · PostgreSQL 17

Intelligent Ticket Triage, SLA Escalation & Auto-Routing System
  • Engineered an automated multi-tier support routing backend in Java 25 and Spring Boot 3, implementing a keyword-based NLP triage engine for instant category classification, severity resolution, and customer priority detection.
  • Designed a real-time load-balancing assignment engine across L1, L2, and L3 engineering tiers based on active workload capacity ratios (active tickets / max capacity), with automated priority upgrade triggers upon SLA escalation.
  • Architected a persistent relational data model in PostgreSQL 17 via Spring Data JPA and HikariCP, capturing full lifecycle audit logs across tickets, ticket history, and transactional email logs with non-destructive database seeding.
  • Built a dynamic React 19 & Material-UI dashboard connected via an Axios API service with automatic mock-fallback resilience for offline testing and seamless RESTful state synchronization.
GitHub Repository ↗
public class TicketTriageEngine {
public RoutingResult triageTicket(Ticket t) {
Category cat = nlpClassifier.predict(t.text());
double ratio = activeTickets / maxCapacity;
if (ratio > SLA_THRESHOLD)
priorityUpgrade(t);
return router.assignTier(cat, ratio);
}
}
FIG. 03 SPEC: AI INTERVIEW SIMULATION PLATFORM · 2025

PrepAI Next.js · Groq · Llama 3.3 70B

AI-Powered Interview Simulation & Candidate Evaluation Platform
  • Designed and iterated prompt templates for Llama 3.3 70B (via Groq) to generate context-aware interview questions and adaptive follow-ups conditioned on job descriptions and prior candidate responses.
  • Built a 5-metric evaluation framework (technical accuracy, communication, business impact, confidence, structured thinking) translating qualitative LLM output into a structured 0–10 STAR-method rubric with detailed feedback.
  • Shipped on a full-stack architecture (Next.js, Node.js/Express) with Groq-hosted Llama 3.3 70B inference, generating detailed, actionable performance reports per candidate.
Live Deployment ↗ GitHub Profile ↗
// ADAPTIVE STAR INTERVIEW GENERATOR
function buildPrompt(jobDesc, history) {
return groq.llama33_70b.generate({
system: "Adaptive STAR Technical Examiner",
context: jobDesc,
priorResponses: history,
rubric: "5-metric 0-10 STAR scoring"
});
}
FIG. 04 SPEC: EVENT-DRIVEN GITHUB AUDIT & DURABLE QUEUE ENGINE · 2026

Scrutinize TypeScript · Node.js · Probot · SQLite · Turso

Event-Driven GitHub Audit & Durable Queue Engine
  • Engineered an event-driven GitHub App in TypeScript/Node.js (Probot) to audit PR review thoroughness; decoupled incoming webhooks into a durable job queue, achieving under 50ms fast-ack responses to eliminate GitHub webhook timeouts.
  • Designed an idempotent queue architecture using SQLite UNIQUE header constraints to silently drop duplicate deliveries, paired with atomic row-level claims (UPDATE ... WHERE status IN ('pending','failed')) to prevent race conditions during concurrent worker sweeps.
  • Developed a deterministic rule engine calculating diff-based reading speed floors (0.4s per line, 30s minimum) and detecting AI co-author signatures; utilized Octokit pagination to bypass default 30-item API caps across PR commits, reviews, and inline comments.
  • Implemented a periodic self-healing sweep daemon for automatic crash recovery (resumes jobs stuck processing over 5 minutes or failed under 5 attempts) and migrated storage to Turso Cloud SQLite (@libsql/client) to prevent data loss on ephemeral PaaS containers.
GitHub Repository ↗
// Idempotent Queue Claim & Atomic Lock
async function claimNextJob(db) {
return await db.run(
"UPDATE jobs SET status = 'processing', locked_at = ?" +
"WHERE id = (SELECT id FROM jobs WHERE status IN ('pending','failed') LIMIT 1)"
);
}
FIG. 05 SPEC: ENTERPRISE HIPAA-COMPLIANT SAAS · 2025

Medicl Autoencoder · Isolation Forest · Streamlit

Enterprise-Grade HIPAA-Compliant Healthcare SaaS Platform
  • Automated missing value handling, outlier detection, and schema validation for healthcare data ingestion pipelines.
  • Integrated anomaly detection using Autoencoders and Isolation Forests to surface irregular patient records and data integrity violations.
  • Built a CLI and Streamlit-based visualization interface for interactive data quality reporting and operational monitoring.
GitHub Profile ↗
class AnomalyDetector:
def __init__(self):
self.ae = Autoencoder(latent_dim=16)
self.iso = IsolationForest(contamination=0.05)
def detect(self, df):
recon_err = self.ae.reconstruction_error(df)
return self.iso.fit_predict(recon_err)
FIG. 06 SPEC: TIME-SERIES FORECASTING SYSTEM · 2024

AI Stock Price Prediction LSTM · GRU · Transformer · FastAPI

AI-Powered Stock Price Prediction System
  • Built end-to-end time-series forecasting using LSTM, GRU, and Transformer models for multi-horizon stock price prediction.
  • Implemented backtesting, cross-validation, and ensemble modeling to improve generalization and reduce overfitting on financial data.
  • Deployed inference using FastAPI + Docker with monitoring for production-grade reliability.
GitHub Profile ↗
class EnsembleForecaster:
models = [LSTM(), GRU(), Transformer()]
def predict(self, X, horizon=30):
preds = [m.forward(X) for m in self.models]
return torch.stack(preds).mean(dim=0)
ACADEMIC & CAREER TIMELINE

The Engineering Journey

01
PASSED 2021
Education

Secondary School Certificate (SSLC)

Elgi Matric Higher Secondary School, Coimbatore — Score: 100%

Completed general academic board curriculum with distinction in Science and Mathematics, securing a perfect 100% score.

02
PASSED 2023
Education

Higher Secondary School (HSC)

Elgi Matric Higher Secondary School, Coimbatore — Score: 86.17%

Completed Higher Secondary Education with focus on Mathematics, Physics, Chemistry, and Computer Science.

03
OCT 2023 — FEB 2024
Internship

Graphic Designer Intern

Visaithalam Solutions

Delivered branding, visual identity, and marketing creatives across client projects under tight deadline constraints, developing visual storytelling and design systems proficiency.

04
2023 — 2027
Education

B.Tech in Artificial Intelligence and Data Science

Dr. N.G.P. Institute of Technology, Coimbatore — CGPA: 8.65 (Till Date)

Specializing in Machine Learning algorithms, Deep Learning architectures, Data Structures & Algorithms, Database Management Systems, and Enterprise Backend Engineering.

05
JUNE 2025
Internship

Cloud Intern

Accent Techno Soft · Coimbatore, India

Gained practical industry exposure in cloud engineering and architecture. Developed hands-on proficiency with Google Cloud Platform (GCP) infrastructure services (Compute Engine, Cloud Storage, IAM, VPC networking) and core Amazon Web Services (AWS) components (EC2, S3, IAM roles), focusing on cloud resource management, containerized deployments, and cloud security policies.

VERIFIED CREDENTIALS

Certifications & Awards

AI Fluency Framework
Anthropic (2025)
VERIFIED
Generative AI Fundamentals
Databricks (2026)
VERIFIED
Elements of AI
University of Helsinki (2026)
VERIFIED
GenAI Powered Data Analytics
Tata (2026)
VERIFIED
LEADERSHIP & ENGAGEMENT

Co-Curricular Activities

Slack Agent Builder Challenge — StalePR

Participated in the Slack Agent Builder Challenge, developing StalePR, an AI-powered GitHub pull request management assistant.

HACKATHON

LinkedIn Newsletter Author — "The AI Edge"

Launched and author the monthly LinkedIn newsletter The AI Edge, sharing insights on artificial intelligence, emerging technologies, and industry trends.

PUBLICATION

Active Member — Robotics Club

Active member of the Robotics Club, participating in technical activities and collaborative engineering projects.

CLUB

Research Paper Presentation — CIT Coimbatore

Presented a research paper on an AI-powered Healthcare Data Handling Platform at Coimbatore Institute of Technology (CIT), Coimbatore.

RESEARCH

Research Paper Presentation — KPR Institute

Presented a research paper on Solar Technology at KPR Institute of Engineering and Technology, Coimbatore.

RESEARCH

Illuminate Entrepreneurship Bootcamp — E-Cell IIT Bombay

Participated in the Illuminate Entrepreneurship Bootcamp organized by the E-Cell, IIT Bombay, focusing on innovation, startup ideation, and entrepreneurship.

BOOTCAMP

Let's Build Something Together

Open to AI/ML engineering roles, Spring Boot backend opportunities, research collaborations, and tech discussions.

Email Me ↗ GitHub ↗ LinkedIn ↗
ESC to exit
Tech Stack Matrix
Jump
Shipped Work & Projects
Jump
Engineering Journey
Jump
Credentials & Certs
Jump
Impact & Engineering Log
Jump
Contact Information
Jump