Open to full-stack & AI engineering roles
Heidelberg · --:-- CET/CEST
Hello, I'm
Full-stack & AI engineer · M.Sc. Artificial Intelligence · Heidelberg
I turn business problems into working software.
Full-stack engineer with 3+ years of experience and a Master's student in Artificial Intelligence. I like starting from the people who will use a product, and I'm increasingly building with generative and agentic AI.
Three projects built end to end, each with source code and a way to see it running.
A used-car dealer platform: a public catalog with full-text search and lead capture, plus a role-based back office for working the lead pipeline. TypeScript from front end to API.
Turns clinical notes into structured summaries that a clinician reviews and approves. Personal data is removed before the model sees anything, and every field cites its source text. Demo only, synthetic data, not a medical device.
Run it yourself: make up
Question answering over technical manuals (PDF and DOCX, German and English) with source citations. The agent grades its own retrieval, rewrites weak queries and checks that the answer is grounded before replying, and it is measured against a naive baseline.
Run it yourself: docker compose up
The LangGraph graph from my Agentic RAG repo, with its real node names. Scroll to follow one example route: a weak first retrieval, a rewritten query, then a grounded answer with citations.
01 · route_question
Is the question about the uploaded manuals? Yes, so it goes to retrieval. Anything else gets an out-of-scope reply.
02 · retrieve
Fetch the closest chunks from the vector store.
03 · grade_documents
Grade each retrieved chunk. This time too few are relevant.
04 · rewrite_query
Rewrite the question and search again, at most twice.
05 · retrieve
Second retrieval with the rewritten query.
06 · grade_documents
Enough relevant chunks now.
07 · generate
Write the answer from those chunks only.
08 · check_hallucination
Check that the answer is grounded in the sources. If not, regenerate once, then say there is not enough information.
09 · format_answer
Return the answer with its citations.
Loops are bounded: at most 2 rewrites and 1 regeneration, and a test proves the graph terminates.
What the projects above have in common, each point taken from the repositories' own READMEs.
Carvio runs CI with an 80% coverage gate and Playwright browser tests. Clinical has 247 automated checks, with 93% line coverage in the API, 87% in the worker and 90% in the web app.
Carvio: every route needs a login unless marked public, roles are enforced on the server, refresh tokens are single-use. Clinical: personal data is removed before the model, a strict content-security policy, an append-only audit trail.
Clinical is evaluated on 33 synthetic notes; the RAG project on 43 questions against a naive baseline, including where it loses. Both READMEs state their limits.
Docker throughout. Clinical adds a Helm chart that ran on a local Kubernetes cluster, plus Terraform and ArgoCD definitions that are validated but not applied to AWS, and says so.
I'm a Master's student in Artificial Intelligence at BTU, based in Heidelberg. Before that I studied Applied Computer Science and worked as a full-stack developer in India and Germany, mostly with React, TypeScript, Java and Spring Boot.
I enjoy the part of the job that happens before the code: sitting with stakeholders, understanding what hurts, and turning it into something people actually want to use. Then I follow it all the way to deployment.
Lately that includes generative and agentic AI. I build proof-of-concepts around semantic search, RAG and workflow automation, always asking whether they solve a real problem.
Oct 2023 – Mar 2025
EGENIE STORES · Remote
Aug 2022 – Jul 2023
Kulturbrauerei Heidelberg · Heidelberg, Germany
May 2019 – Jul 2021
Adhaan Solution Pvt. Ltd. · Ahmedabad, India
M.Sc. Artificial Intelligence
Brandenburg University of Technology (BTU)
Artificial intelligence, machine learning, data analytics, intelligent systems.
Since Apr 2025
M.Sc. Applied Computer Science
SRH Hochschule Heidelberg
Enterprise computing, process management, data strategy, UI/UX. Erasmus+ scholar in Austria, 2023.
Oct 2021 – Sep 2023
B.Tech Computer Science
Maharshi Dayanand University, Rohtak
May 2015 – Jun 2019
The portfolio is a project too. These are facts about the code you are looking at.
I'm looking for a full-stack or AI engineering role where I can build things that matter. If that sounds like your team, I'd love to hear from you.
© 2026 Vansh · Designed and built by Vansh with Next.js, Tailwind CSS, Motion and three.js.
Back to top ↑