AI Systems & Agentic Workflows
AI Systems Builder
I build the infrastructure layer that makes businesses run on AI — end-to-end pipelines, LLM integrations, and automated data flows that replace manual work. Currently embedded in Finance & Operations at Tanso Technologies, where I also own the AI infrastructure.
My background spans data science, financial modeling, and consulting. I'm curious about where AI genuinely changes how businesses operate — from ops and finance to sales intelligence.
Always up for a conversation about AI architecture, applied LLMs, or climate tech.
Current role
Finance & Operations Manager
Tanso Technologies
Feb 2026 – Present · Munich
Work History
- Built custom data pipeline integrating 3 SaaS systems (HubSpot, Personio, DATEV) into a real-time reporting layer serving C-suite and external VC/angel investors; deployed on Azure Container Apps with Entra ID auth
- Engineered meeting intelligence platform on GCP processing 1,200+ calls since March 2026 (300+/month): webhook-driven ingestion, pgvector embeddings, and autonomous Claude agents extracting structured insights
- Deploying agentic automation across 4 business functions: Sales, Customer Success, Marketing, and Operations
- Built financial models and KPI dashboards for management reporting and operational planning
- Coordinated vendor evaluations and tool onboarding as Tanso scaled its operational infrastructure from Seed stage
- Automated key operational workflows and implemented business tooling to support Tanso's scaling from Seed stage
- Financial analysis and damages quantification across 5+ international arbitration cases in automotive (billion-euro dispute), offshore wind, and consulting
- Built financial models and valuations across energy, infrastructure, and financial services
- Python and Excel for large-dataset analysis; outputs fed into expert witness reports
- Built financial models and business cases for wind energy and CDR technology investments
- Market analysis covering competitive dynamics and investment trends in European climate tech
- Developed Power BI dashboards and VBA automation tools for internal reporting
- IT and internal control audits with SAP focus
- System migration audits
- Taught statistics and maths to 30–50 undergraduate students per semester
Academic Background
CFA Level 1
CFA Institute
MSc Business Administration & Data Science
Copenhagen Business School
BSc International Business & Entrepreneurship
Leuphana University
Toolkit
AI / Agents
Data Science
Finance
Languages
Selected Work
Knowledge Layer
Unified agent-accessible knowledge layer connecting HubSpot, Intercom, and product usage data into a single queryable system. Structured for LLM consumption — normalized schemas, semantic retrieval, and tool interfaces for Claude agents operating across Sales, CS, and Marketing.
Meeting Intelligence Platform
Webhook-driven meeting intelligence platform on GCP. A custom connector streams calls into Cloud Run, where autonomous Claude agents extract structured insights. Transcripts are vectorised with pgvector and surfaced via a custom MCP server for semantic search over every sales call.
Investor Reporting Dashboard
Custom Python data pipeline connecting HubSpot, Personio, and DATEV APIs into a unified reporting layer. ETL processes normalize and consolidate data from 3 systems; deployed on Azure Container Apps with Entra ID auth and role-based access control.
FinWave
Local speech-to-text for macOS. Records from the menu bar, transcribes via Whisper (MLX) fully on-device, with optional Claude cleanup. Paste into any app instantly — no audio leaves the machine.
View on GitHubFinFlow
macOS menu-bar speed reader using RSVP (Rapid Serial Visual Presentation). Feed it a PDF or Markdown and it flashes one word at a time with ORP (Optimal Recognition Point) highlighting — your eye stays fixed while words flow past. Vim-style keyboard control, chapter navigation.
View on GitHubMSc Thesis: ML in Wind Energy
Gaussian Process and SVR metamodels for load case selection in wind turbine certification at Stiesdal. Multi-sensor approach that minimises expensive physical simulations without sacrificing accuracy. Grade: 1.0.
Stiesdal CDR Dashboard
Interactive Plotly/Dash dashboard for the carbon-dioxide-removal (CDR) certificate market, built at Stiesdal Sky Clean. Turns live cdr.fyi transaction data into filterable views of purchases, deliveries, top suppliers and buyers, and price trends across CDR methods (biochar, DAC, enhanced weathering).
View on GitHubReview Topic Modelling
Unsupervised topic modelling of ~82k Amazon reviews of Apple products, comparing Top2Vec and BERTopic — including a from-scratch BERTopic reimplementation (UMAP → HDBSCAN → c-TF-IDF). Surfaces real themes like battery life, carrier unlocking, and cases/accessories. CBS NLP final project.
View on GitHubMars Landmark Classification
Deep-learning classification of NASA HiRISE Mars surface landmarks across 8 terrain classes, using AlexNet and GoogLeNet CNNs with an SVM baseline. Investigates whether imbalance handling (ADASYN / Random Oversampling) improves accuracy on a severely imbalanced dataset — finding that architecture matters more than rebalancing. CBS MSc group project.
View on GitHubCFA Level I
CFA Institute
Feb 2025
MSc Business Administration & Data Science
Copenhagen Business School
1.1 · Top 1%
BSc International Business & Entrepreneurship
Leuphana University
1.3 · Top 5%
Get in Touch
Open to conversations about AI, finance, and climate tech.