Hello, I'm
Avanish Singh
AI Engineer
Building production-grade generative AI systems — RAG pipelines, LLM applications, and applied machine learning.
About01
I'm a final-year AI student at Technische Hochschule Deggendorf (Grade: 1.7), with hands-on industry experience shipping production generative AI systems.
At Swisscom, I built a GPT-4o RAG pipeline that reduced hallucinations by 40% and a real-time voice-to-avatar assistant with sub-400 ms latency. I'm currently contributing to AI-assisted research tooling at Medizinische Hochschule Hannover, and previously worked as an ML Engineer at KOgenie, where I deployed predictive models that improved forecasting accuracy by 25%.
I'm looking for a full-time AI Engineer role — particularly in LLM applications, RAG architecture, or applied ML for real-world products.
Experience02
Student Research Assistant
Medizinische Hochschule Hannover (MHH)
- Developed software tooling to streamline data workflows for the U-WaTCH pediatric pulmonology research project.
- Collaborated with Prof. Dr. Christine Happle and Frank Mueller on interdisciplinary research across programming, data entry, and application development.
AI Engineer — Academic Partnership
Swisscom
- Built a GPT-4o RAG pipeline (LangChain + pgvector) that reduced hallucinations by 40% compared to Swisscom's baseline FAQ bot.
- Prototyped a voice-enabled 3D customer service avatar achieving <400 ms speech-to-render latency using WebRTC and Unreal Engine MetaHuman.
- Deployed Dockerised FastAPI and Next.js microservices with CI/CD via GitHub Actions and Supabase Edge Functions.
Machine Learning Engineer
KOgenie E-solutions
- Designed and deployed predictive models (Random Forest, Neural Networks) that improved forecasting accuracy by 25%.
- Built automated Python/SQL data pipelines, reducing processing time by 40% and operational costs by 15%.
Featured Projects03
View All ProjectsGenAI 3D Avatar — Swisscom
Voice-to-avatar customer service assistant for Swisscom's Digital CX team. GPT-4o RAG pipeline (LangChain + pgvector) reduced hallucinations by 40%. Real-time lip-sync via Unreal Engine MetaHuman over WebRTC achieved sub-400 ms latency, autonomously handling 78% of Tier-1 queries.
Emotion Detection — LSTM from Scratch
End-to-end NLP pipeline classifying 20K+ tweets into six emotion categories. Built a deep LSTM with embedding layer, stacked LSTMs, and dropout regularisation from scratch — achieving 0.92 F1-score. Validated with confusion matrices and t-SNE embedding visualisations.
AdversEco — AI Ad Creation SaaS
SaaS platform that automates digital ad creation using OpenAI generative models, with OAuth-integrated direct publishing to Instagram, LinkedIn, Twitter, and Facebook. Built with real-time scheduling and a Next.js frontend.
Skills04
AI & Machine Learning
Engineering & Infrastructure
Application Development
Education05
B.Sc. Artificial Intelligence
Technische Hochschule Deggendorf
Grade: 1.7 (German scale) · Python, Data Science, NLP, Machine Learning
Get In Touch06
I'm actively looking for full-time AI Engineer roles. If you're working on interesting problems with LLMs, RAG, or applied ML — let's talk.