AI engineering · Capstone
A mortgage assistant with cited answers, validated calculations and an application the customer controls.
Top 3 in my AI Engineering cohort
Product engineer building AI systems · Stockholm
A product engineer, going deeper into AI
I’ve spent nine years building software across different industries, including finance, healthcare and operations. My recent projects focus on AI, from retrieval and evaluation to the interface people use.
- 9 yearsbuilding product software
- Top 3AI Engineering capstone
- 8 peopleuse Paisa across 4 households
Projects
AI systems and everyday software.
Product engineering · Finance
The app my family uses to plan the month, split bills and see what is left.
8 people · 4 households · In daily use
AI engineering · Evaluation
A lab where I compare embeddings, chunking and retrieval methods, then use recall and latency to weigh the trade-offs.
195 retrieval configurations compared
AI engineering · Local models
A reply-drafting assistant I’m building for a friend’s clinic. Every draft needs his review before it can be sent.
44/44 development evaluation cases · Local 14B model
About me
From product software
to AI systems.
I’m Kartick, a frontend engineer at Lendo in Stockholm. I studied applied physics and electrical engineering before moving into product software.
My favorite part is still the moment something works for the first time.
Since March 2026 I’ve been building LLM systems end to end. I want people to be able to check the source behind an answer and understand when the system cannot help.
At Lendo I also trace production issues through our Go services. Outside work it is mostly family and football.
How I work with coding agents
I’ve built a harness that coordinates coding agents for planning (Claude Opus 5.5), orchestration and code review (Claude Sonnet 5.5), implementation (Claude Haiku 5.5 and Gemini 3.8 Flash), and local commit and PR drafting (Qwen3 Coder 30B). I review the changes and run tests.
I’m running model routing evals to compare a stronger implementation model with a stronger planner followed by cheaper models for implementation and checking plan adherence. I measure task success, review and revision cycles, and cost per accepted task to test whether fewer iterations offset higher model costs.
My goal is to run everything from implementation to PR creation entirely locally.
Reflections
Making local AI easier to use.
During 2026, I’ve been thinking about how people could run open-source AI agents locally and use them to automate parts of their work.
The part that interests me is how to connect those agents into a workflow people can understand and shape themselves. I’d like to make that easier for people who don’t work in AI.
Other projects I’m proud of
MinForskning
A digital consent platform for healthcare research. I designed the patient flows in Figma and led the frontend. The client asked me back in 2024 for the next phase.
Lead frontend and UX/UI · AFRY for Uppsala Universityminforskning.seHertz Nordic
A fleet platform used across Sweden, Norway and Finland. I led the frontend architecture, moved legacy Angular apps to React, set up testing and Azure CI/CD, and helped rebuild the public Hertz Freerider booking site.
Lead frontend · AFRYFreerider booking siteBAS-X
A two-person SAAB hackathon project for planning aircraft turnaround. Users can simulate a plan, replay it, branch at a decision point and compare outcomes.
Frontend lead · SAAB hackathonCode Demo