Dhanush Chandra ShekarDhanush C.

MS Data Science at Indiana University, graduating 2026. I build end-to-end intelligent systems — from training Transformer models in PyTorch to shipping production MLOps pipelines and multi-agent LLM architectures.

[ active parameter ]
hover to explore
Q
W
E
R
T
Y
U
I
O
P
A
S
D
F
G
H
J
K
L
Z
X
C
V
B
N
M
selected work

Featured Projects

Production-ready architectures bridging LLMs, vector embedding indexes, MLOps orchestration, and structured analytics models. All repositories are open source on GitHub.

01

LLM Gateway: Unified API Orchestration

PROBLEM

Managing OpenAI, Anthropic, and Gemini separately created fragmented infra with inconsistent latency and no fault isolation.

ACTION

Built a unified FastAPI gateway with semantic caching, intelligent routing, and circuit breakers behind a single endpoint.

RESULT

98% reduction in p95 latency with zero-downtime failover and full Prometheus observability.

LLM Gateway: Unified API Orchestration
FASTAPIREDISPOSTGRESQLPROMETHEUSDOCKER
02

OrgMind: Multi Agent State Engine

PROBLEM

Teams using Slack, Notion, and GitHub had no shared memory — context was siloed and data conflicts went undetected.

ACTION

Engineered a 12-agent LangGraph system backed by Neo4j to manage organizational state and flag conflicting transactions in real time.

RESULT

Cross-platform state resolution across 3 integrated environments with automatic conflict detection.

OrgMind: Multi Agent State Engine
LANGGRAPHNEO4JGROQCHROMADBREACT
03

LodeAI: Intelligent Recruitment Platform

PROBLEM

Technical hiring relied on manual code reviews, creating bottlenecks and inconsistent candidate evaluation at scale.

ACTION

Built an AI-driven platform with VS Code integration, Docker-sandboxed code execution, and automated evaluation pipelines via Claude AI.

RESULT

Full recruitment workflow from submission to structured scoring, fully automated end to end.

LodeAI: Intelligent Recruitment Platform
NEXT.JSTYPESCRIPTSUPABASECLAUDE AIDOCKER
04

GraphRAG: Knowledge Retrieval System

PROBLEM

Standard vector search over SEC filings missed multi-hop relationships between entities, reducing answer quality for complex queries.

ACTION

Combined Neo4j graph traversal with vector embeddings to enable entity-aware, multi-hop Q&A retrieval over financial documents.

RESULT

Outperformed vector-only baseline on context extraction across a 46-query benchmark suite.

GraphRAG: Knowledge Retrieval System
NEO4JCHROMADBCLAUDE APISENTENCE TRANSFORMERS
05

Transformer: Neural Machine Translator

PROBLEM

Off-the-shelf NMT models are opaque — hard to study attention or customize training without abstraction layers in the way.

ACTION

Built a complete encoder-decoder Transformer from scratch in PyTorch with custom attention, positional encoding, and training pipeline.

RESULT

Fully functional sequence-to-sequence translation model trained end-to-end with zero external NMT dependencies.

Transformer: Neural Machine Translator
PYTORCHPYTHONNLPCUDA
06

Demand Forecasting: Production MLOps Pipeline

PROBLEM

Retail demand forecasts degraded silently over time — no retraining triggers, no drift monitoring, no deployment automation.

ACTION

Built a full MLOps pipeline with XGBoost/LightGBM ensemble, automated retraining, drift detection, and CI/CD via GitHub Actions.

RESULT

MAE of 0.2575 with zero-touch deployment and continuous monitoring in production.

Demand Forecasting: Production MLOps Pipeline
XGBOOSTLIGHTGBMFASTAPIDOCKERGITHUB ACTIONS
07

Mental Health NLP: Discourse Classifier

PROBLEM

Mental health discourse on Reddit is nuanced and class-imbalanced — generic classifiers fail to categorize it reliably.

ACTION

Fine-tuned DistilBERT with NLPAug data augmentation to handle severe class imbalance and improve cross-category robustness.

RESULT

72.5% F1 accuracy on a highly imbalanced multi-class mental health classification task.

Mental Health NLP: Discourse Classifier
HUGGINGFACEDISTILBERTSCIKIT LEARNNLPAUG
08

InsightFlow: Star Schema ETL Pipeline

PROBLEM

Raw transactional data had no dimensional structure — BI reporting was slow and analytical queries were hard to maintain.

ACTION

Designed a star schema and built a production ETL pipeline with PostgreSQL and SQLAlchemy, surfaced via Power BI dashboards.

RESULT

Analytics-ready data layer enabling self-serve BI with structured dimensional queries across the full dataset.

InsightFlow: Star Schema ETL Pipeline
PYTHONPOSTGRESQLSQLALCHEMYPOWER BIETL
09

A/B Testing: Production Framework

PROBLEM

Marketing decisions were based on noisy experiment data — no rigorous statistical framework to validate conversion lifts.

ACTION

Built an automated A/B testing framework with scipy statistical testing, ETL reporting cycles, and Tableau dashboards.

RESULT

12% conversion rate lift validated through statistically significant testing with automated reporting.

A/B Testing: Production Framework
PYTHONSCIPYSTATSMODELSPOSTGRESQLTABLEAU
tech stack

Skills

Languages
PythonTypeScriptSQLC
AI / ML
PyTorchLangGraphHuggingFaceTransformersRAGXGBoostLightGBMscikit learn
Data & Databases
Neo4jChromaDBFAISSPostgreSQLRedisSQLAlchemy
Frameworks
FastAPINext.jsReactNode.jsLangChain
DevOps & Tools
DockerGitHub ActionsPrometheusSupabasePower BITableau
background

Experience

2024 to PresentBloomington, IN

Faculty Assistant in Data Science

Indiana University · Kelley School of Business

Graduate students struggled to bridge statistical theory and production-grade code. Redesigned lab workflows around real datasets, authored reproducible Python exercises for 30+ students, and built rubrics that graded engineering practice alongside math.

TeachingData SciencePythonStatistics
2024 to 2026Bloomington, IN

MS in Data Science

Indiana University Bloomington

Coursework alone doesn't produce production-ready engineers. Pushed every project toward real deployments — building graph databases, vector search pipelines, and MLOps systems that went beyond academic exercises into working open-source software.

MLNLPNeo4jMLOpsPyTorch
get in touch

Contact

Available for full time roles, research collaborations, or intelligent systems consultation. Let's build together.

LOCATIONBloomington, Indiana (USA)
STATUSOPEN TO FULL TIME OPPORTUNITIES
SYSTEM SYNCHRONIZEDLOADING...
SEND MESSAGE
hover keyboard to explore