Open to Junior ML Engineer & Data Scientist roles · Berlin / EU

I build AI systems
and measure them.

AI Engineer & Data Scientist with an MSc in Data Science and AI. I take projects from messy data to deployed apps: LLM agents, retrieval-augmented chatbots and forecasting models, each with tests, CI and an honest evaluation.

Portrait of Dipanshu Singh
3+ yrsenterprise experience in B2B SaaS
25+Qlik Sense BI dashboards built
70%of reporting workflows automated
3live demos you can try now
How I work

Every project follows the same loop.

A model is only as good as the evidence behind it. I build the evaluation before I trust the result, and ship it where people can try it.

01

Data

Clean, version and split it so the test set never leaks into training.

pandas · SQL · Open-Meteo · MedQuAD
02

Model

Start from a simple baseline, then earn every bit of added complexity.

scikit-learn · Keras · Prophet · e5 · LLMs
03

Evaluate

Held-out benchmarks, confidence intervals and failure cases, reported as measured.

rolling-origin · hit@k · bootstrap CI
04

Ship

Tested code, CI on every push, and a demo anyone can open.

pytest · GitHub Actions · Streamlit · Docker

↺ measure, learn, iterate

Selected work

Projects

Six projects across AI, machine learning and data science. Each links to its code; three run live.

RAG · Healthcare● live

Medical RAG Chatbot

Answers health questions from 16,407 NIH Q&A pairs, cites its sources, and refuses questions outside its knowledge base.

Right condition in the top 3 for 86% of lay-language test questions (TF-IDF baseline: 43%); refuses 17 of 20 off-topic ones.

  • e5-base
  • FAISS
  • Groq LLM
  • Streamlit
  • pytest + CI
LLM agent · Text-to-SQL

DataAnalyst AI Agent

Connect a database or upload CSV/Excel, ask in plain English, and get SQL, tables and charts back.

Optional double-check mode flags when two answers disagree, measured on a public Text-to-SQL benchmark.

  • LLM agents
  • SQL
  • Docker
  • uv
  • CI
Time series · Deep learning

Berlin Temperature Forecasting

7-day forecasts comparing a Bidirectional LSTM, a Transformer and Prophet against naive baselines, under leakage-controlled rolling-origin evaluation.

BiLSTM best at 3.10 °C MAE over 44 weekly forecasts, but for tomorrow alone, "same as today" still wins.

  • TensorFlow/Keras
  • Prophet
  • Open-Meteo
  • 19 tests + CI
Text-to-SQL · Retrieval

Olist Analytics Agent

A Text-to-SQL agent over the Olist Brazilian e-commerce dataset, with retrieval, evaluation and a Streamlit UI.

  • LLMs
  • SQL
  • Retrieval
  • Streamlit
NLP · Labour market● live

AI Job Automation Risk

NLP analysis of 1,000+ job descriptions, scoring how exposed each role's tasks and skills are to AI automation.

  • spaCy
  • NLP
  • Streamlit
Regression● live

House Price Predictor

Predicts house prices in major Indian cities with an XGBoost regression pipeline behind a Streamlit front end.

  • XGBoost
  • scikit-learn
  • Streamlit

More on GitHub, including multi-modal music genre classification and PM2.5 forecasting with LSTMs.

Skills & background

Comfortable across business and engineering.

Languages
Python, SQL
ML
scikit-learn, TensorFlow/Keras, Prophet, XGBoost
NLP & LLMs
RAG, embeddings, FAISS, LangChain, spaCy, agents
Shipping
FastAPI, Streamlit, Docker, GitHub Actions, pytest
Analytics
pandas, Qlik Sense, Tableau, Power BI, Matplotlib
  • MSc Data Science and AI, Berlin.
  • 3+ years in B2B SaaS: built 25+ Qlik Sense dashboards and automated 70% of reporting workflows.
  • Business impact: insights that contributed to a 15% quarterly revenue uplift and 20% churn reduction across clients including Acer, Apple, Godrej and Reliance.
  • 40+ enterprise implementations led, growing product usage by ~10% quarter on quarter.
  • Learning now: MLOps, model deployment and LLM evaluation.
Contact

Let's talk.

Hiring for a junior ML or data science role in Berlin or the EU? I'd like to hear from you.