AI & NLP Engineer — MSc Data Science

Building
production
AI systems

MSc Data Science from the University of Europe for Applied Sciences, Potsdam. I research and deploy RAG systems, LLM pipelines, multi-agent architectures, and cloud infrastructure. 8 live deployed systems.

GitHub → HuggingFace LinkedIn

rohithkumar336699@gmail.com

+49 17665131866

Berlin, Germany — Job Seeker Visa — EU Blue Card Eligible

RAG System

93%

Recall@10 on 8.84M MS MARCO passages.
Optimal alpha=0.70 — validated p=0.002

LLM Fine-Tuning

0.089%

Parameters trained with QLoRA.
Loss 2.47 → 0.89 on free Colab T4

LLM Evaluation

0.877

RAGAS overall score. Faithfulness 0.909.
Gemini as judge LLM — 10 test questions

Portfolio

8 live

Deployed AI systems across RAG, agents,
fine-tuning, cloud, Kubernetes, MLOps

Hybrid RAG — 93% Recall@10 —

QLoRA Fine-Tuning — TinyLlama 1.1B —

LangGraph Multi-Agent — 4 Specialised Agents —

AWS EC2 — systemd Auto-Restart —

Kubernetes — Rolling Updates —

RAGAS Evaluation — 0.877 Overall —

MLflow + DagsHub — Public Dashboard —

MSc Data Science — Berlin 2026 —

Hybrid RAG — 93% Recall@10 —

QLoRA Fine-Tuning — TinyLlama 1.1B —

LangGraph Multi-Agent — 4 Specialised Agents —

AWS EC2 — systemd Auto-Restart —

Kubernetes — Rolling Updates —

RAGAS Evaluation — 0.877 Overall —

MLflow + DagsHub — Public Dashboard —

MSc Data Science — Berlin 2026 —

01 — About

From thesis
to production

My MSc thesis discovered that alpha=0.70 — not the commonly assumed 50/50 split — is the optimal BM25/dense fusion weight for hybrid retrieval. Validated with paired t-test p=0.002. The thesis research contributes to work published on arXiv (2507.18910).

After graduating I did not stop at the thesis. I built and deployed 8 production AI systems — fine-tuning LLMs with QLoRA, building 4-agent LangGraph pipelines, deploying to AWS EC2 with systemd, orchestrating with Kubernetes, and tracking experiments with MLflow.

I am actively seeking ML Engineer, NLP Engineer, and AI Engineer roles in Berlin. Job seeker visa. EU Blue Card eligible. Available immediately.

Core ML & NLP

PythonPyTorch HuggingFaceRAG FAISSBM25 E5-base-v2

LLM Stack

LangChainLangGraph QLoRARAGAS GeminiUnsloth

Infrastructure

DockerKubernetes AWS EC2FastAPI systemdUbuntu

MLOps

MLflowDagsHub HuggingFace SpacesGit

02 — Projects

8 live systems,
all deployed

→

01 — Thesis

Hybrid RAG System

BM25 + E5 dense embeddings on 8.84M MS MARCO passages. Alpha=0.70 discovered through 11 tracked experiments.

93% Recall@10 — +11.4% over baseline — MRR=1.0

FAISSBM25E5FastAPIDocker

Demo API GitHub

→

02 — Agent

ReAct AI Agent

LangGraph ReAct agent with 3 tools — web search, calculator, RAG retrieval. Decides which tool to use based on the question.

3 tools — LangGraph state graph

LangGraphGeminiDuckDuckGoStreamlit

Demo GitHub

→

03 — Fine-Tuning

QLoRA Fine-Tuning

TinyLlama 1.1B on NLP domain knowledge. 4-bit quantisation + LoRA adapters. Free Colab T4 GPU. Published on HuggingFace Hub.

0.089% parameters — Loss 2.47→0.89

QLoRAUnslothTinyLlamaHuggingFace

Model GitHub

→

04 — Evaluation

LLM Evaluation

RAGAS + Gemini as judge LLM. 5 metrics across 10 NLP test questions. Live Streamlit dashboard with radar chart.

Faithfulness 0.909 — Overall 0.877

RAGASGeminiStreamlitPlotly

Dashboard GitHub

→

05 — Cloud

AWS EC2 Deployment

FastAPI RAG inference API on EC2 t3.micro Frankfurt. systemd auto-restart on failure and server reboot. SSH key auth.

eu-central-1 — 3 REST endpoints — OpenAPI

AWS EC2UbuntusystemdFastAPI

Live API GitHub

→

06 — Orchestration

Kubernetes

FastAPI RAG API — 2 replicas, RollingUpdate, liveness and readiness health probes, resource limits. Zero downtime deployments.

2 replicas — RollingUpdate — Health probes

KubernetesDockerkubectlYAML

GitHub

→

07 — Multi-Agent

Multi-Agent Research

4 LangGraph agents — Search (DuckDuckGo), Summarise (temp 0.3), Fact-Check (temp 0.1), Writer (temp 0.4). Each agent specialised for its task.

4 agents — LangGraph state graph

LangGraphGeminiDuckDuckGoDocker

Demo GitHub

→

08 — MLOps

MLflow Tracking

3 experiments tracked — RAG alpha search (11 runs), QLoRA loss curve (16 steps), RAGAS evaluation. Public dashboard on DagsHub.

11 alpha runs — Public MLflow dashboard

MLflowDagsHubPython

Dashboard GitHub

03 — Results

Numbers that matter

RAG Retrieval

93%

Recall@10 on 8.84M passages
MS MARCO benchmark

Statistical Validation

p=.002

Paired t-test significance
alpha=0.70 validated

Fine-Tuning

0.089%

Parameters trained with QLoRA
Loss reduced 64%

LLM Evaluation

0.909

Faithfulness score
RAGAS evaluation framework

Portfolio

8 sys

Live deployed AI systems
All public and documented

Improvement

+11.4%

Over BM25 baseline
Hybrid retrieval advantage

04 — Education

Background

Mar 2024 — Mar 2026

MSc Data Science

University of Europe for Applied Sciences — Potsdam, Germany

Thesis: Hybrid RAG combining BM25 sparse retrieval with Microsoft E5-base-v2 dense embeddings. 93% Recall@10 on 8.84M MS MARCO passages. Optimal alpha=0.70 validated with paired t-test p=0.002. Thesis research contributes to work on arXiv (2507.18910).

2026 — Present

Independent AI Engineer

Berlin, Germany — Job Seeker Visa

Built and deployed 8 production AI systems covering RAG, LLM fine-tuning, multi-agent architectures, LLM evaluation, AWS EC2, Kubernetes, and MLflow experiment tracking. All systems publicly accessible and documented on GitHub.

05 — Contact

Let us work
together

● Available immediately — Berlin

Ready to build
real AI systems
with your team

Actively seeking ML Engineer, NLP Engineer, and AI Engineer roles in Berlin. Job seeker visa holder. EU Blue Card eligible. I bring 8 deployed systems and deep hands-on experience debugging production ML infrastructure.

Phone

+49 17665131866

rohithkumar336699@gmail.com

rohith-kumar-reddipogula

GitHub

RohithkumarReddipogula

HuggingFace

Rohith2026

MLflow Dashboard

DagsHub — Public experiments