Building ARIA · Brain-Inspired AI, Fraunhofer IEM
Habeeb
Sanni
Generative AI engineer building multimodal LLM systems, RAG pipelines, and industrial AI automation.
About
21
AUTONOMOUS AGENTS — ARIA’s three-tier cognitive architecture at Fraunhofer IEM
4
LLMs ORCHESTRATED — GPT-4, LLaMA, Mistral, DeepSeek at Echocene
I'm an M.Eng. student specializing in Generative AI, multimodal LLM systems, and industrial AI automation. I build data pipelines on Azure, fine-tune open-source LLMs, and develop RAG-based systems with LangChain — backed by several years of hands-on experience in performance marketing and technical sales.
Currently completing my Master’s thesis at Fraunhofer IEM — ARIA, a brain-inspired multi-agent cognitive architecture that is self-healing and self-improving — after an LLM/ML engineering internship at Echocene. My marketing background in Google/Meta Ads and SEO still shapes how I think about shipping AI products people actually use.
Ollama
Local-First Cognition in ARIA
Azure
Bronze / Silver / Gold Pipelines
n8n
Workflow Automation
SEO
Growth Marketing Background
Technical Focus
Where AI engineering depth meets a marketer’s instinct for what actually gets used.
GENAI & LLM
- LangChain
- Ollama
- GPT-4
- LLaMA
- Mistral
- DeepSeek
- FastAPI
- Whisper
DATA & MLOPS
- Azure
- Bronze/Silver/Gold
- Data Lineage
- SQLite
AUTOMATION
- LangChain Agents
- n8n
MARKETING & GROWTH
- Google Ads
- Meta Ads
- SEO
- WordPress
Featured Projects
Brain-Inspired Multi-Agent AI · M.Eng. Thesis
ARIA
A self-healing, 21-agent cognitive architecture that diagnoses its own failures and patches its own source code.
Multi-Vault RAG System
eshu
Local-first multi-vault RAG system that routes questions to the right knowledge base and flags contradicting sources.
German Language Learning
LexiDesk
Turns any textbook PDF into a self-updating, spaced-repetition vocabulary widget — fully offline, no cloud API calls.
Experience
- Built a multimodal RAG pipeline orchestrating GPT-4, LLaMA, Mistral and DeepSeek for regulatory-compliance queries in construction (CSRD, GEG, EU Taxonomy).
- Designing ARIA, a brain-inspired multi-agent cognitive architecture that decomposes perception, memory, reasoning, and metacognition into 21 specialised agents across three tiers: Core Pipeline, Adaptive Layer, and Evolutionary Layer.
- Built a self-healing lifecycle: the system detects its own failures, diagnoses and drafts a source-code patch, sandbox-tests it, and automatically rolls back on regression.
- Runs local-first on open-weight LLMs via Ollama for everyday cognition, reserving a cloud model exclusively for the single highest-stakes operation — modifying its own source code.
- Built a Bronze/Silver/Gold Azure data pipeline and LangChain RAG agents that cut manual effort in data engineering and reporting.
Get in
Touch
I care about AI systems that are actually used — grounded in real data pipelines, evaluated honestly, and explainable to the people who rely on them.
📍 Köthen, Saxony-Anhalt — open to relocating anywhere
✉️
💼
🐙
GitHub
📍
Based In
Köthen, Germany · Open to Relocate