Bhavesh Mahender Jain
GenAI, multi-agent systems, and strategic foresight. Building AI that turns evidence into decisions.
01About
I am a data scientist and Team Lead Data & Development at the Joint Innovation Hub, working at Fraunhofer ISI. My work connects generative AI, multi-agent systems, and strategic foresight: turning research signals into tools that help people make decisions.
I am building Vorschau and work on the JIH Trend Radar. I also publish research on multi-agent retrieval and transparent AI systems, and write about evaluation and AI product development.
English · basic German
My Fraunhofer ISI profileTap to turn the card
02Tools & methods
My working stack and the AI methods I study. Explore an element.
03Selected work
Current foresight platforms and multi-agent research, followed by three new project concepts.
A multi-agent foresight platform for small and mid-sized companies. Agents scan signals, weigh evidence, and turn research into briefings for strategic decisions.
6+ years in data science and 12+ production systems shipped, across research and industry.
The in-development platform coordinates 7 agents across 12k+ sources, with under 4 minutes from question to briefing.
Technology intelligence used by innovation teams across the EU public-sector network.
My M.Sc. work studied demand forecasting for air cargo, with 12% improvement against an ARIMA baseline.
Three focused ideas for reliable agents, evidence-led foresight, and AI product decisions. Proposed projects; no results claimed.
Researched 4 October 2026A workbench that asks whether an agent completed the task—not just whether its final answer sounded right. Replay tool calls, inspect state changes, and compare reliable completion across repeated runs.
Why it fitsBuilds on my research in transparent multi-agent systems and my focus on LLM evaluation.
Start with a synthetic logistics booking workflow, a sandbox database, and a small frozen task set. Compare a single agent with a verifier-assisted workflow.
Proposed stackPython · SQLite · pytest · OpenTelemetry · Interactive web UI
Compare final database state with the expected result; report repeated-run task success, policy violations, attack success, latency, and cost separately.
A technology-scouting workspace where every trend claim has an evidence trail. Connect research signals, expose contradictory findings, and show how a conclusion changes when evidence is added or removed.
Why it fitsExtends my technology-readiness and bibliometric research with a sharper focus on evidence quality.
Use a frozen, openly available corpus for one technology area. Build a claim ledger and source-backed briefing, then evaluate on questions held out before tuning.
Proposed stackPython · OpenAlex · PostgreSQL · GraphRAG baseline · Interactive web UI
Compare graph retrieval with keyword and vector baselines using citation validity, evidence coverage, contradiction recall, reviewer usefulness, and cost. Research momentum is not a validated technology-readiness score.
A product-discovery workspace that turns an AI feature idea into a testable decision: a user task, a simple baseline, a cost and latency budget, and evidence for a release or a stop.
Why it fitsConnects my AI product learning with my writing on discovery, evaluation, and the work around a demo.
Build one synthetic research-briefing use case. Compare manual/search and retrieval-assisted workflows, document failure modes, and generate a release brief with explicit go, revise, or stop criteria.
Proposed stackPython · FastAPI · PostgreSQL · LLM evaluation · Interactive web UI
Compare decisions against a reviewed rubric on fixed product scenarios. Measure evidence coverage, task quality, cost, latency, and reviewer agreement. A readiness checklist is decision support, not proof that a product will succeed.
04Ongoing learning
The questions and methods I am studying across AI systems and product development.
Grounding useful outputs in the right evidence.
Choosing when tools and autonomy improve a workflow.
Making coordination visible, testable, and worth its cost.
Turning a promising capability into a useful product.
05Writing
Notes on AI systems, useful products, and the decisions around both.
06The path so far
Selected experience, with the most recent work first.
04/2024 — Present
Fraunhofer ISI · Joint Innovation Hub
Currently Team Lead Data & Development. Building GenAI and foresight tools, including the JIH Trend Radar.
05/2022 — 03/2024
Fraunhofer IML
Air-cargo forecasting, NLP, and machine learning research.
07/2022 — 06/2023
SAP
Analytics Cloud dashboards and ML-supported planning.
10/2021 — 03/2022
TU Dortmund
Data mining research and Big Data Analytics tutoring.
08/2019 — 12/2019
Amazon India
Python automation and dataset validation.
M.Sc. Data ScienceUniversity of Potsdam
B.Tech. Computer Science EngineeringSreenidhi Institute of Science and Technology
07Research & output
Concrete work and output from my current portfolio, with sources you can follow.
Jain, B. M., Shaji, G., & Kumar, D.
HCI International 2026 · Lecture Notes in Computer Science · Springer
DOI: 10.1007/978-3-032-29586-6_24Jain, B., & Kumar, D.
ISPIM Conference · 15–18 June 2025
DOI: 10.24406/publica-5777Kumar, D., & Jain, B.
EPIA 2025 · Lecture Notes in Computer Science · Springer
DOI: 10.1007/978-3-032-05179-0_22with Busch, M., Kumar, D., Valentowitsch, J., & Wurster, D.
19th Symposium for Foresight and Technology Planning · Berlin, October 2025
DOI: 10.17619/UNIPB/1-246708Start a conversation
Python