The iPhone’s LiDAR sensor measures the real volume of the food on your plate, so calories start from a measurement instead of a guess.
My own algorithm models your body with a Kalman filter and retunes your calorie target every week.
Scan a barcode or read a nutrition label, and the food is logged in seconds.
Jacob Soendergaard
I build AI agents, automation and native iPhone apps.
Two apps, designed and built by me
It reads your records and can change them through tool calls, and every change can be undone.
Log a dose in a tap and see every logged, missed and planned dose on one calendar.
Each dose comes out of the right vial, so you always know how long your supply lasts.
What I do
- PepNote’s assistant edits real records through tool calls, with undo on every action and a spend cap per account
- Custom MCP servers and agent skills for Claude Code, Cursor and Codex, including one that lets agents reuse shared web components
- Hermes Agent and OpenClaw set up for marketing and support, with subagents and model routing
- Support agents for client businesses that answer incoming enquiries and take new customers from first question to a quote, so staff skip the routine questions
- Swift and SwiftUI throughout, from onboarding to subscriptions, with PepNote submitted to the App Store
- A LiDAR meal scanner that finds the plate’s plane, builds a height map and turns depth into food volume
- iOS App Foundry, my own Swift packages for sign-in, subscriptions and account deletion
- Supabase and PostgreSQL with row-level security and sync across devices
- Edge Functions for streaming AI chat, meal scans, RevenueCat webhooks and account deletion
- Food lookups cut from 28 ms to 2.9 ms
- A Kalman filter tracks each body, and a GRU network trained on 2 million simulated people refines the calorie target
- Trained in PyTorch with domain randomisation and a DAgger round, then ported to native Swift to run on the iPhone
- Multimodal meal scans with structured outputs, on a model chain that falls back automatically
- LoRA fine-tuning of FLUX for images and Qwen for more consistent RAG answers
- Every screen of CaloMatic and PepNote, from Figma to SwiftUI, including onboarding, paywalls and App Store screenshots
- A 550-icon food library for CaloMatic, built to one style guide
- This site: my name lit on my real GitHub contribution graph in custom 5×7 lettering, and every scene drawn live on a 32×32 cell display
838 contributions
On GitHub in the last 12 months, mostly in private repositories.
Experience
CaloMatic
2026–nowAI calorie tracking, in development. Photo recognition on multimodal LLMs, and my own adaptive calorie algorithm, tested in simulation on noisy and missing logs.
PepNote
2026Peptide and GLP-1 tracking. Designed and built in SwiftUI, then taken through App Store submission.
Agent systems
2026An MCP server and agent skill that let coding agents reuse shared web components. Hermes Agent and OpenClaw for marketing and support.
Model fine-tuning
2025–2026FLUX for image generation, and Qwen with LoRA for more consistent RAG answers.
Freelance AI and web development
2024–nowClient websites built with AI-assisted development and generative visuals. Support agents that answer incoming enquiries and move new customers from first question to a quote faster, so staff no longer spend their time on routine questions.
Aalborg University Copenhagen
2025–nowBSc in Computer Science, 3rd semester, expected 2028. Algorithms, data structures and web programming.
NEXT Sukkertoppen
2022–2025HTX, technical upper-secondary, Computer Science and Geoscience.
