Open to Werkstudent, internship & full-time AI/ML roles in Germany
Hi, my name is
I spent two years shipping enterprise backends and now work on GenAI full time: RAG pipelines, eval harnesses, and Infrastructure-as-Code. Based in Berlin, Germany, work permitted.
Prototype fast. Ship to production. Measure everything.
From enterprise Java in Bangalore to LLM research in Berlin.
My path didn't start with AI. It started with two years of enterprise Java/Spring Boot at Vrize in Bangalore. Real production systems for real clients, where a bad deploy meant angry users, not a failed notebook cell. That's where I learned to care about testing, CI/CD, and code that other people can maintain.
Then LLMs changed what software could do, and I wanted in, not as a spectator. So I moved to Berlin for an M.Sc. in Big Data & AI at SRH (graduating Sept 2026), where I also work as a Research Assistant building end-to-end RAG pipelines, ChromaDB retrieval, chunking and embedding tuning, and structured extraction.
Today I'm focused on LLM evaluation, retrieval, and GenAI features that survive contact with real users, and my master's thesis is an empirical study of verifier-guided LLM self-repair across three frontier models. The result is a rare combination: I build AI products with backend-engineer discipline: sandboxed execution, eval harnesses, streaming APIs, and Terraform-managed infrastructure. Prototypes are easy. Things that stay up are not.
From enterprise sprints in Bangalore to AI research in Berlin.
Sovrano AI· Remote
SRH University of Berlin· Berlin, Germany
Vrize India Pvt Ltd· Bangalore, India
Vrize India Pvt Ltd· Bangalore, India
07/2024 – Expected 09/2026 · Berlin, Germany
SRH University of Berlin
2018 – 2022 · Bangalore, India
BMS College of Engineering, Bangalore
Built to ship, not just to demo. Each one solves a real problem end to end.
Production-deployed analytics platform with a natural-language-to-SQL pipeline, sandboxed query execution, real-time SSE streaming, and auto-generated Recharts visualizations. Includes a Gaussian Copula synthetic-data engine with fidelity metrics and a text-to-SQL eval harness with execution-accuracy benchmarking tracked in MLflow.
A SaaS platform that replaces live meetings. Teams submit input asynchronously through a guided chat powered by Claude Sonnet, which asks follow-up questions on its own, then synthesizes all responses into a structured decision summary on a real-time WebSocket dashboard.
An Infrastructure-as-Code pipeline that provisions complete AWS accounts from a single YAML config: IAM users, groups and policies, S3 buckets, and VPC networking, all through Terraform and GitHub Actions with OIDC auth. It also writes a per-user runbook after every deploy.
AWS Account Onboarding Automation
An empirical study of whether LLMs can self-repair the Terraform they write. A pipeline generates Terraform, checks it against four independent referees (syntax/plan, security via Checkov + tfsec, org policy via OPA/Conftest, and a completeness check), then feeds the errors back for up to 10 repair rounds: 560 runs across 50 scenarios, 3 seeds, and three models, logging 2,068 errors down to the rule and round. The headline finding is a capability ladder, not a single aggregate number: Claude Sonnet 5 resolves 95% of tasks and fails only on unfamiliar resource types, GPT-4o resolves 53% and fails specifically on cross-resource reasoning like networking, and Llama 3.1 resolves 9% and fails to converge at all. A controlled experiment isolating feedback quality as the only variable shows detailed error messages causally drive repair success (McNemar p < 0.0001), and a repair-integrity classifier that diffs code before and after each fix found the frontier models never cheat by deleting a required resource, only the weakest model does. Reproduced against the IaC-Eval (NeurIPS 2024) benchmark: the one-shot GPT-4o score (21%) matches their published 19.36%, and repair lifts it to 45%.
Master's Thesis
Beyond the featured work
Academic and personal builds across computer vision, NLP, and developer tooling. Some are in private beta ahead of release.
Remote MCP server on Cloudflare Workers that lets Claude on my phone browse my repos and file ideas as issues. OAuth 2.1 with PKCE, a server-side PAT (no token passthrough), and a deliberately read-heavy toolset: nine tools, one write.
Chess improvement app that pulls your chess.com games, runs Stockfish in the browser to classify blunders by game phase, and coaches you through your weak spots with talking pieces.
CNN- and GAN-based models that flag AI-manipulated images, separating authentic artwork from synthetic forgeries.
Semantic enrichment pipeline that surfaces under-described or less-popular videos using NLP and Google Cloud AI APIs.
Text-based language identification for Indian languages using a Naive Bayes classifier over n-gram features.
COVID-era biomedical waste segregation using an ANN with VGG16 transfer learning for image classification.
The tools I reach for to take an idea from notebook to production.
Cloud and AI badges from Google Cloud and AWS Educate, each publicly verifiable.
Peer-reviewed research, presented at an international conference.
Conference Paper
Ohrid Conference, via SRH University of Berlin
A stacking ensemble model that combines LSTM and Random Forest predictions in an XGBoost meta-learner to explore how supplier stock movements relate to Tesla's stock price.
Hiring, collaborating, or just curious about a project? Say hi.
RAG pipelines, prompt engineering, eval harnesses, and Claude API integrations that make it to production.
FastAPI and Spring Boot services, RESTful APIs, microservices, and real-time WebSocket/SSE systems.
Terraform, AWS/GCP/Azure, GitHub Actions with OIDC, Docker, and CI/CD pipelines.
Document classification, information extraction, analytics, and synthetic data with fidelity metrics.
