Rudraksh Sharma

Applied AI Engineer & Qiskit Advocate

Data Scientist at AIONOS building production RAG systems, LLM applications, and hybrid quantum-classical optimization solutions. Published researcher with 7 papers and international hackathon recognition.

My Background

About Me

Data Scientist at AIONOS and Technical Lead at Beerantum, working across LLM engineering and quantum optimization. My day-to-day is building production RAG systems: embedding pipelines, hybrid retrieval (BM25, FAISS, Qdrant), and advanced rerankers including cross-encoders and BGE. KnowledgeOS, my flagship open-source project, spans the full retrieval stack from raw ingestion to benchmarked agentic retrieval.

On the quantum side, I work on hybrid quantum-classical optimization for logistics, aviation, and supply chain problems, with published research and industry work through the QuBri Initiative. My Physics background (MSc, NSUT) shapes how I think about both domains. Optimization at scale is an energy-landscape problem regardless of whether the solver is a QUBO on annealing hardware or a loss surface in PyTorch. The interesting engineering is always in the composition.

Core Skills

AI / LLM Engineering
Python LLM Engineering RAG LangChain FAISS ChromaDB Qdrant Sentence Transformers FastAPI MLflow Docker Deep Learning NLP scikit-learn SQL
Quantum Computing
Qiskit PennyLane D-Wave Ocean Quantum ML QUBO VQE QAOA Quantum Annealing Statistical Analysis

Recognition & Awards

Qiskit Fall Fest 2025

Winner

QIntern 2025

First Team Award

Berlin Kipu Quantum Hackathon 2026

3rd Place Winner

Featured Work

Projects

Production systems, research code, and applied AI across LLM engineering and quantum optimization.

LLM Engineering / MLOps

KnowledgeOS: Production RAG Platform

A modular, benchmark-driven retrieval platform built from first principles across 6 milestones. Covers the complete pipeline: 5+ embedding strategies, hybrid indexing (BM25, FAISS, Qdrant, RAPTOR), and 8 retrieval strategies including multi-hop, Self-RAG, CRAG, and agentic ReAct. Benchmarked end-to-end: CrossEncoder reranker matches agentic recall at 72x lower latency.

Python FAISS Qdrant ChromaDB BGE Reranker Sentence Transformers Ollama OpenRouter

recall@1 = 1.000  |  CrossEncoder at 190ms/query  |  788x L1 cache speedup

LLM Engineering

Finetune-Pipeline

End-to-end domain SLM fine-tuning platform. Point at any document corpus, generate synthetic training data, fine-tune via QLoRA on Kaggle, export to GGUF, deploy with Ollama. Switch clients by editing one config file.

QLoRA Unsloth GGUF Ollama HuggingFace
LLM Security

Sentinel-AI

7-layer adversarial defense framework for LLM applications. Detects indirect prompt injection, token smuggling, sandbox escapes, social engineering, and RAG exploitation. 50-test suite, 0% false positives on legitimate requests.

FastAPI Embeddings Python

90% detection rate  |  0% false positives

Agentic AI / LangChain

AgentPitch: Agentic RFP Automation

End-to-end agentic proposal automation for IT services firms. Auto-analyzes RFPs and generates tailored proposals. 68/68 tests passing, FastAPI backend with a production-ready UI.

LangChain Groq LLMs FAISS FastAPI

68/68 tests passing

Applied AI / NLP

Clinical-Doc-AI

Context-aware, explainable clinical NLP system for automated medical document understanding. Designed around real-world healthcare workflows with full explainability for regulatory alignment.

NLP SHAP Python scikit-learn
Quantum / ML

Quantum Water Pollutant Classifier

Hybrid Quantum-Classical Neural Network identifying contaminant ions in water using experimental photonic sensor data. Linked to published paper (Jun 2026).

Qiskit PennyLane PyTorch

96.67% accuracy on IBM quantum hardware

Quantum / Finance

Quantum-Classical Stock Forecaster

Hybrid Quantum-Classical RNN (LSTM + VQC) for stock index closing price prediction, leveraging a variational quantum circuit for enhanced feature learning.

LSTM VQC PennyLane PyTorch
Research

Publications

7 publications across quantum optimization, hybrid quantum-classical systems, and quantum hardware experiments.

Article Jul 2026

Quantum-Inspired Optimization for Robust Aircraft Cargo Loading Using a QUBO Formulation

Rudraksh Sharma, Ravi Katukam, Arjun Nagulapally

Reformulates aircraft cargo loading as a QUBO problem encoding payload maximization and structural safety. Classical heuristics produce fragile solutions near operational limits; the quantum-inspired approach yields robust placements under perturbation.

Read Paper
Preprint Jan 2026

Bridging the Linear-Quadratic Gap: A Quantum-Classical Hybrid Approach to Robust Supply Chain Design

Rudraksh Sharma, Ravi Katukam, Arjun Nagulapally

Addresses the trade-off between demand coverage and facility overlap in urban logistics. Tested on a high-fidelity Delhi NCR dataset; hybrid quantum-classical solver outperforms classical optimization on the Pareto frontier of demand capture vs. overlap risk.

Read Paper
Preprint Dec 2025 arXiv:2512.23009

Symmetry-Preserving Variational Quantum Simulation of the Heisenberg Spin Chain on Noisy Quantum Hardware

Rudraksh Sharma

Investigates physics-informed ansatz design for VQE on NISQ devices. Symmetry-preserving circuits outperform hardware-efficient baselines on IQM Garnet hardware, validated against exact diagonalization.

Read Paper

Get in Touch

Open to discussing new projects, research collaboration, or freelance opportunities.