CHIRAG NATESH VIJAY GERMANY

AI Engineer
× Venture
Builder

I turn signals into systems. Explainable ML at Bosch, agentic AI and RAG pipelines on GitHub, and Cellvara, my AI decision intelligence startup for pharma and biotech teams.

Agentic AIRAGLLM AppsMachine Learning Explainable AIGenAIMLOpsAI Product
SCROLL

WHO I AM

Research depth.
Product instinct.
Founder energy.

I'm an AI engineer and founder based in Germany. Over the last six years I've shipped machine learning in industry at Bosch eBike Systems, done signal processing research at the University of Passau, published IoT work with ACM, and patented an agriculture drone back home in Bengaluru.

An AI system should verify its own output, explain itself, and stay honest. Confidence without evidence isn't intelligence.

These days I split my week between applied AI work at Hilti, a second master's at Neu-Ulm, and building Cellvara. The hardest problems in AI right now aren't purely technical. They're about trust, workflows, and getting real teams to actually adopt this stuff.

CURRENT VENTURE

Cellvara

Pharma and biotech SMEs know AI could help them. Most just can't afford six-figure consultants to figure out where, how, and whether it's safe. Cellvara is an AI consulting copilot that turns that uncertainty into a structured answer: the right use case, the evidence behind it, and a compliance-aware path to implementation.

Assessment → Workspace → Documents → Consultant Chat → Report → Collective Learning
Next.js 14TypeScriptConvex Multi-model LLMsDoc intelligencePDF reports Voice I/O

Grounded, not generic

Consultant chat that reasons over the company's own profile, uploaded documents and reusable organisational memory. Not vibes.

Built for regulated reality

Explainable steps, traceable reasoning and human-in-the-loop governance, designed around R&D, QA, regulatory and pharmacovigilance workflows.

Forged in the ecosystem

Shaped through customer discovery with pharma professionals, the Cyber Valley AI Incubator and UnternehmerTUM / TUM Startup Launchpad.

SELECTED WORK

Systems that check
their own work.

Twelve builds across agentic AI, RAG, multimodal systems and applied ML. Every card says what it really is: open source, prototype, research or hackathon build.

SHOWING 12 / 12
Open sourceRAG

DocChat

Ask a question, get an answer you can trust. A multi-agent RAG system where verification agents fact-check every response against the source documents before it reaches you. Hybrid BM25 + vector retrieval with self-correction loops.

DEMONSTRATESReliable RAG & verification architecture

LangChainVector DBBM25Verification agents
PrototypeAgentic

PrepAI

Drop in your CV and a job description, get a personalised mock interview and a coach-style report. Evaluator and coach agents score your answers on clarity, depth and relevance through a contract-first API.

DEMONSTRATESAgent orchestration, structured outputs & evaluation

LangGraphFastAPIWhisperGradio
Research demoSafety

MedAssist MAS

A healthcare triage assistant that knows its limits. Specialist agents run the consultation while deterministic red-flag checks catch emergencies and force escalation. The safety logic never depends on the model behaving.

DEMONSTRATESSafety-aware agents & structured generation

AutoGenPydanticGradio

Educational demo, not a medical device

HackathonAgentic

AgnesTheSecond

Supply-chain intelligence for CPG teams, built in one weekend at the TUM.ai × Spherecast Makeathon 2026. A function-calling agent queries live procurement data, finds ingredient substitutes and explains every tool call across 1,000+ products and 40 suppliers.

DEMONSTRATESConversational analytics over a real data model

FlaskOpenAI functionsSQLiteGraph views
PrototypeVenture AI

ForgeFoundry

An AI co-founder for the messy zero-to-one phase. Multi-agent pipelines take a raw idea and produce problem framing, market research, ICP personas, business models, go-to-market strategy and a generated pitch deck.

DEMONSTRATESEnd-to-end multi-agent product pipelines

Multi-agentVenture strategyPitch generation
Open sourceSpeech

Speech Analyzer AI

Record a meeting, walk away with the minutes. Transcribes speech, cleans up messy domain jargon, and turns an hour of talking into structured summaries and action items.

DEMONSTRATESSpeech AI, domain adaptation & workflow automation

WhisperIBM watsonxLangChain
Open sourceMultimodal

AI NourishBot

Point it at a photo of your plate. Specialist agents identify the ingredients, break down the nutrition and suggest recipes. Vision-language AI applied to something everyone does three times a day.

DEMONSTRATESMultimodal AI & coordinated specialist agents

CrewAIVisionwatsonx
Open sourceRAG

VidIntel

A three-hour YouTube lecture, answered in seconds. Extracts the transcript, summarises it, and answers questions grounded strictly in what was actually said. Its limitations are documented in the README, not hidden.

DEMONSTRATESVideo intelligence & retrieval pipelines

FAISSwatsonxRAGYouTube
Open sourceAgentic

Job Applying Buddy

Feed it a vacancy, your CV and your GitHub profile. Role-based CrewAI agents research the position, tailor your story to it, and generate polished application materials as ready-to-send PDFs.

DEMONSTRATESDocument automation & role-based agent teamwork

CrewAIPDF generationGradio
Architecture specApplied ML

Banking Fraud Detector

What would it actually take to run fraud detection in production? A full system spec engineered around latency, recall, throughput, drift monitoring and security. The unglamorous constraints that decide whether ML survives contact with reality.

DEMONSTRATESProduction ML thinking & measurable system design

KafkaMLOpsDrift detectionmTLS / PII
ResearchVision

COVID-19 X-Ray Detection

Deep residual networks classifying chest X-rays across COVID-19, pneumonia and healthy cases. Transfer learning in a domain where careful evaluation and error analysis genuinely matter.

DEMONSTRATESMedical imaging & transfer learning

ResNet-18PyTorchMedical imaging
Open sourceCreative AI

Jazz Music Generation

An LSTM that learned jazz. Trained on chord and note sequences, it improvises original solos one note at a time. Early proof that sequence models can swing, and that AI can be playful too.

DEMONSTRATESGenerative sequence modelling

LSTMKerasMusic AI

THE PATH

Bangalore → Passau →
Bosch → Founder.

2026 — Present
Working Student, Artificial Intelligence
Hilti Group · Kaufering, Germany
Applied AI in an industrial context. Translating business workflows and data into practical prototypes and implementation paths.
2025 — Present
Founder
Cellvara (formerly ZestLearn) · Germany
Product strategy, AI architecture and customer discovery for responsible AI adoption in pharma & biotech. Selected for the Cyber Valley AI Incubator and TUM Startup Launchpad.
2026
Impact Strategy Consultant
Snowball Effect · via SEA:lab, Social Entrepreneurship Akademie
Redesigned impact measurement for a social enterprise: theory of change, a KPI dictionary and a credible data pipeline from survey to decision.
2023 — 2025
Data Science & Machine Learning
Bosch eBike Systems · Kusterdingen, Germany
Intern → thesis → working student. Rebuilt industrial ETL from hours to minutes across 100,000+ sensor records, automated acoustic reporting (~80% less manual effort), and delivered an explainable vibration→sound model with <5% MAE.
2022 — 2023
Research Assistant
FORWISS · University of Passau
Benchmarked NetCDF/HDF5 and compression approaches for large wavelet-signal datasets: data models, I/O throughput and decode latency.
2019 — 2021
Undergraduate Research Assistant
Centre for IoT · PES University, Bengaluru
Built a modular IoT retrofit platform for legacy buildings with facial-recognition access, smart HVAC and leak detection. Published at ACM VISP.

RESEARCH & IP

Peer-reviewed.
Patented. Explainable.

Master's thesis

Predicting Perceived E-Bike Motor Loudness & Sound Quality from Vibration Data with Explainable AI

Explainable Boosting Machines that translate vibration signals into third-octave acoustic bands and perceived loudness at under 5% MAE, with global and local interpretability that engineers can act on. University of Passau × Bosch eBike Systems, 2024.

XAI · EBM · SIGNAL PROCESSING
Publication

A System to Retrofit Existing Infrastructure to Be Smart and IoT Ready

Modular smart-infrastructure architecture for retrofitting homes and campuses with low-cost edge devices and MQTT/REST orchestration. Published at the 4th International Conference on Vision, Image and Signal Processing (ACM VISP).

DOI: 10.1145/3448823.3448864 ↗
Patent

A Flying Object for Dispensing Chemical Composition and Method Thereof

Autonomous precision-agriculture drone using CNN-based computer vision to detect diseased plants and dispense treatment only where it's needed. Less blanket spraying, better resource use. Indian Patent Office, application 201941047691.

COMPUTER VISION · UAV · AGRITECH
M.Sc. Int. Entrepreneurship, Digitalization & Sustainability
Hochschule Neu-Ulm
2026 — PRESENT
M.Sc. Computer Science — Artificial Intelligence
University of Passau
2021 — 2024
B.Tech. Computer Science & Engineering, with distinction
PES University, Bengaluru
2016 — 2020

TOOLBOX

Agentic & Generative AI

LangGraphLangChainCrewAI AutoGenLlamaIndexMCP RAGStructured outputsAgent evaluation OpenAI · Gemini · watsonx

ML, Data & Signals

PythonPyTorchPandas OptunaInterpretML / EBMExplainable AI Signal processingETL pipelinesMySQL

Product Engineering

Next.jsReactTypeScript TailwindFastAPIFlask ConvexVercel

Venture & Impact

Customer discoveryProduct strategy Responsible AI adoptionTheory of change KPI designEN C1 · DE B1/B2 · HI/KN native

IN THEIR WORDS

What teammates say.

Pulled straight from LinkedIn. People I've led, shipped with and studied beside.

An exceptional leader who balances strategic vision with hands-on support. He kept the team focused, motivated and aligned, and knows exactly how to bring out the best in a team.

Chetachi Okereke
Senior Account Manager · Industrial Equipment & Workplace Safety
Worked with Chirag on the same team

His ETL pipelines slashed data preprocessing from hours to just 5–10 minutes, and his improvements to our Explainable AI sound-rating model raised accuracy by roughly 55%. Technically strong, proactive and a great collaborator.

Aseem Garg
Data Scientist · ML & AI Specialist
Chirag's team at Bosch

Exceptional expertise in data science, responsible machine learning and AI. Beyond his technical strengths, he is an outstanding and reliable team player, always willing to help colleagues.

Anju Koyippilly Rathnan
Software & Data Engineer
Worked with Chirag at Bosch

He used Explainable AI to rate the sounds of different e-bike drive units, making complex models understandable and useful. Above all, he is incredibly talented and a great team player.

Vinodhini R
Data Analyst & BI Expert
Worked with Chirag at Bosch

Impressive problem-solving skills and deep expertise in Python, machine learning, deep learning and IoT. What I admire most is his willingness to patiently help others until everyone is on the same page.

Ashwin Siddharth
Cloud Specialist Engineer · HPE Aruba
Studied with Chirag at PES University

Proficient across Python, machine learning, deep learning and IoT. His problem-solving abilities were a valuable asset to the outcome of every project we worked on together.

Ankit Pathange
Business Intelligence Specialist
Studied with Chirag at PES University

WANT THE FULL STORY?

This site is the demo.
Chirag OS is the deep dive.

chirag-os · visitor session
$ whoami
chirag: AI engineer, founder, human. This page is just the trailer.
$ man chirag
Full documentation lives in Chirag OS: interactive project teardowns, the decision logs, certifications, case studies and the thinking behind every build.
Booting… the deep dive is one click away.

Away from the terminal

SAY HELLO

Building something
in AI? Let's talk.

Hiring for AI or ML roles? Curious about a Cellvara pilot? Or just want to talk agents, RAG and startups? My inbox is open.

or write to chiragatgermany@gmail.com