Hunch
Giving AI agents a memory worth keeping.
TECH LEAD & FULL STACK ENGINEER
Serious code.Human touch.
I’m David. I build the systems behind the screen — and care just as much about the people in front of it.
Leading engineering at YLM Solutions.
Creating Hunch and what comes next.
SCROLL TO EXPLORETHOUGHTFUL BY DESIGN. ENGINEERED TO WORK.
End-to-end ownershipFrom schema to production
Open-source mindsetBuilding with a wider community
Curiosity, appliedAI agents to local inference
THE INVENTOR’S WORKBENCH
Every part.
A purpose.
Good software is a system of small things working together. Pull this one apart.
One product. Interface, services, and data working as one.
01 / Interface02 / Services03 / Data
SELECTED WORK
Built with intent.
Production systems. Open-source tools.
A few of the problems I’ve made my own.
Soldi
From the database to the people in the field.
Coding Orchestrator
The right agent. The right task. Evidence attached.
Local Whisper Transcriber
Your voice, transcribed on your own machine.
A feel for the front end.
A few interface explorations, too.
THE SYSTEMS ARCHITECT
Making the invisible
work beautifully.
THE PERSON BEHIND THE PULL REQUEST
I connect
the dots.
Between the product and the platform.
The big picture and the tiny detail.
The team and the thing we’re building.
I came through QA, customer support, and project management before moving into engineering. That path still shapes how I build: understand the problem, question the assumptions, and own the outcome.
Today I lead technical direction at YLM Solutions. Outside work, I build open-source tools and explore what AI can do when the engineering around it is solid.
Meet me in the codeDEPTH, IN MORE THAN ONE DIRECTION
The part that has to work when nobody is watching.
APIs, multi-tenant systems, background workers, and the data models underneath them. From SQL query tuning to the last notification arriving.
Complex systems. Clear experiences.
Responsive interfaces, RTL Hebrew workflows, reporting, maps, and cross-platform field apps. I care about how a product feels as much as how it works.
Beyond the prompt. Into the engineering.
MCP servers, agent orchestration, and local inference research. My experiments span Qwen3 MoE routing in llama.cpp, offline Whisper, and hybrid retrieval with BM25 and rank fusion.
Ownership continues after deployment.
Production releases, cloud infrastructure, independent workers, backups, and disaster recovery. I make architectural decisions with the operating reality in mind.
THE PATH HERE
More perspective.
Better software.
YLM Solutions
2019 — PRESENT
Operations software forservice businesses in Israel.
Tech Lead & Full Stack Engineer
MAR 2026 — PRESENTOwning a multi-tenant SaaS platform from SQL schema to deployment. Leading architecture, modernizing the live Angular app, and building the services and Flutter client around it.
Angular migration with zero downtime. A separate notification service. Stronger backups and lower cloud spend.
Frontend Developer
AUG 2023 — APR 2026Modernized reporting with Vue and JSON, shipped an integrated chat app, and built React tools that take repetitive work off the team’s plate.
Project Manager
SEP 2019 — AUG 2023Turned real customer needs into specifications, partnered with developers, and led QA. Learned to ask better questions before writing the first line.
Before YLM Project management · Support · QA
dooble
Project Manager
JAN 2018 — SEP 2019HaAutomatzia HaHadasha
Customer Support Engineer · Municipal systems
NOV 2016 — JAN 2018Fundtech
QA Engineer · Banking systems
2000 — 2003ALSO IN THE OPEN
Contributing to adrkit : stale-decision warnings, consistent machine-readable output, and Windows reliability.
THE JOURNEY, IN WORDS
Still learning. Still building.
Engineering memory, large local models, and the experiments that surprised me. Findings from the workbench, with the evidence left in.
Local LLMs · Field notes
30 billion parameters. No GPU.
What running Qwen3 on a six-core, 64 GB machine taught me about active parameters, memory bandwidth, and the difference between fitting and running well.
4 min read
Building Hunch
Code remembers what. Who remembers why?
The idea behind Hunch: keep the decisions, failed approaches, and context that an AI coding session should not have to rediscover.
3 min read
Engineering memory · Evidence
Your agent remembered. Did it actually learn?
Building Hunch pushed me to separate four things that are too easily collapsed: recall, claimed use, verification, and actual improvement.
3 min read
WHAT’S NEXT Based in Israel. Thinking beyond it.