DE

I build production AI
in research and industry.

Data scientist and Team Lead Data & Development at the Joint Innovation Hub, by day at Fraunhofer ISI. Six years building machine learning systems — forecasting, LLM evaluation, and strategic foresight. I write about what holds up once it is live.

See the work Read the writing
Based in Heilbronn, DE
Currently Vorschau & Trend Radar
Focus LLM eval · forecasting · foresight
Published 4 peer-reviewed papers
Portrait of Bhavesh Jain
HEILBRONN · DE
Working stack
Python PyTorch LangGraph FastAPI Postgres pgvector Docker AWS · GCP Weights & Biases Pandas · Polars scikit-learn Anthropic · OpenAI APIs
01Selected work

AI systems that work
after the demo.

A short selection. I share outcomes where I am allowed to.

Building · vorschau.ai

Vorschau, foresight for SMEs

A multi-agent platform that helps small and mid-sized companies see what is coming next. Agents scan signals, weigh evidence, and write decision-ready briefings.

7 agentsWorking in sequence
12k+ sourcesPatents, papers, news
< 4 minFrom question to briefing
Fraunhofer ISI

JIH Trend Radar

A tool that spots emerging technologies before they hit the trade press. Bibliometric mining and language models cluster weak signals into clear trend cards.

Used by innovation teams across the EU public-sector network.

M.Sc. thesis

Air cargo forecasting

Demand forecasting for a German logistics company.

12% better than the ARIMA baseline.

Publications

Four peer-reviewed papers

Time series, forecasting, and NLP for strategic foresight.

See publications →

02Publications

Peer-reviewed work.

Conference papers and proceedings on multi-agent systems, retrieval, and technology foresight.

  1. 2026

    From Black Box to Glass Box: Designing Transparent Multi-agent Systems for Enterprise Users

    Jain, B. M., Shaji, G., & Kumar, D.

    Lecture Notes in Computer Science · Springer Nature Switzerland

    DOI: 10.1007/978-3-032-29586-6_24
  2. 2025

    Multi-agent Retrieval-Augmented Generation for Enhancing Answer Generation and Knowledge Retrieval

    Kumar, D., & Jain, B.

    EPIA 2025 · Lecture Notes in Computer Science · Springer

    DOI: 10.1007/978-3-032-05179-0_22
  3. 2025

    Predicting Tech Readiness through Bibliometric Analysis using Unsupervised Machine Learning

    Jain, B., & Kumar, D.

    ISPIM Conference · 15–18 June 2025

    DOI: 10.24406/publica-5777
  4. 2025

    Fallstudienbasierte Analyse der Integration generativer KI-Werkzeuge in die strategische Vorausschau

    with Busch, M., Kumar, D., Valentowitsch, J., & Wurster, D.

    19th Symposium for Foresight and Technology Planning · Berlin, October 2025

    DOI: 10.17619/UNIPB/1-2467
03Track record

The numbers, briefly.

0
Years in data science
Forecasting · NLP · GenAI
0
Peer-reviewed papers
Time-series · foresight · NLP
0
Production systems shipped
Across research & industry
0
Agents in the Vorschau pipeline
Signal scan → briefing
04How I think

Writing, in the open.

Notes on evaluation, retrieval, and foresight — a flagship piece, plus recent writing.

Flagship · Evaluation · 2026-04-28

How to evaluate LLM apps

Most LLM apps ship without a real evaluation set — so quality drifts and nobody notices until users do. A practical way to build evals that catch regressions before they reach production.

Read the essay →
Evaluation · 2026-04-04 The cost of average accuracy RAG · 2026-04-20 RAG is not a silver bullet Foresight · 2026-04-08 Foresight vs. forecasting More on the blog →
05Play

Try the ideas, not just read them.

Six tiny interactive ML demos — built from scratch, no libraries. Click around to build intuition.

Linear regression Thompson sampling K-means Confusion matrix A/B test Bias–variance
Open the playground →
06About

Short version.

I am a data scientist based in Heilbronn, Germany. Most of my work sits where machine learning meets decisions people actually have to make — demand forecasts, technology scouting, evaluation for language-model systems.

At Fraunhofer ISI I work on strategic foresight: bibliometric mining, weak-signal detection, and the JIH Trend Radar. As Team Lead Data & Development at the Joint Innovation Hub I look after the data and engineering side of that work. Alongside it I am building Vorschau, a multi-agent foresight platform for small and mid-sized companies.

Before that: an M.Sc. thesis on air cargo demand forecasting, and five peer-reviewed papers on time series, forecasting, and NLP. I write regularly on the blog, mostly about evaluation — the part of ML that decides whether anything else you did mattered.

Get in touch

Say hello.

Always glad to talk about machine learning, evaluation, or foresight — a question, a paper, or something you are building.

Email me Connect on LinkedIn Résumé
Heilbronn, Germany CET · async friendly English & basic German