Dan Gittik

Observability for Agents

Logika teaches a coding agent to define and manage system events in a central ledger – which means it can record all the relevant schemas and expected behaviors alongside the code. This, in turns, gives the agent a consistent way to instrument production systems, know when something breaks, and fix it almost as soon as it does. What's more, it renders the complicated post-processing done by legacy platforms somewhat redundant, giving you both observability and analytics at a low, flat fee.

Tools for Agents to Improve Their Own Work

Nock started from the idea that a CLI can give coding agents a direct, standardized way to operate on a codebase. Some of this work is generic – as such, it is (or will be) provided by the harness, sooner or later; but most of it is custom to your particular system – so by capturing structured evidence about the its own operation, an agent can reflect on past sessions, draw conclusions, and extend this toolbox as your project evolves, improving speed and accuracy while reducing cost and churn.

Real-time, Adaptive Conversational AI

At UG Labs, I led the development of a platform for real-time AI characters for children. We developed our own streaming conversation library before Live APIs were available, coordinating speech recognition and synthesis, language models and safety guardrails into a responsive voice-to-voice experience with sub-second latency. We combined parallel model execution, LLM-based evaluation and dynamic conversational state when these approaches were still emerging in mainstream AI tooling, as well as custom transcription of children's speech and LoRA-based image generation for branded assets. Across the platform, we worked with 35+ open models and 8+ inference providers, balancing capabilities with performance and reliability.

Agents for Financial Automations

At Apprentice, I developed agents that repair enterprise financial automations when they fail mid-run: navigating and investigating about large workbooks of interconnected formulas, understanding the original intent and fixing discrepancies while the user waited; context engineering, model selection and focused tools brought the process from 10+ minutes down to 1, while improving success rate by ×2-10. I also implemented a headless Excel executor to give agents better access to tinker with automations: it let them play out hypotheses and inspect the consequences at a fraction of the cost and speed, accelerating our end-to-end test suite by ×3-20. To do that, I built evaluation pipelines that measure repair quality and performance, using LLMs to judge agent outputs and additional checks to evaluate the judges themselves.

Cybersecurity and Distributed Systems

I started in R&D at the Ministry of Defense, working on reverse engineering and low-level clients in C and assembly. As the system grew, my work expanded into servers, networking, databases, provisioning and DevOps, then product development and adoption. I led a team of 6 engineers (2 of whom became managers) and helped develop a technical field that grew from 3 people to 50+; honorably discharged as captain.

Android and Machine Learning

At Google London, I was responsible for display, brightness and color in Android, working from Pixel drivers through native code, Binder and Java to the user interface. Alongside systems work, I combined color theory and perception research with supervised and unsupervised machine learning on telemetry from billions of users to improve how screens adapted to people’s preferences and behavior.

Deep Learning and Spatial Computing

At Magic Leap, I designed and built the Memory Graph: a pipeline combining deep learning, computer vision and spatial sensor data to represent objects and their relationships in a user’s surroundings. A knowledge graph and spatial-computing API made that information available to applications, allowing for the development of augmented realty experiences that were unprecedented at the time.

Teaching Engineering through Practice

I joined Tel Aviv University as a teaching assistant and became a lecturer after one semester, revamping its flagship information security course, which I later co-authored for edX, reaching 50,000+ learners in 150+ countries, ranked #1 in cybersecurity and #6 in all of computer science on Class Central (Best Online Courses, Best of All Time, 2020 edX Prize finalist). I also created Advanced System Design, where students built a complete, distributed data-processing system, from the wire protocol to the web UI; and Impact-Oriented Programming, where they developed real solutions for research labs and NGOs over a semester, experiencing first-hand the challenges of communication with clients and the potential of open-source software.

Engineering, Consulting & Teaching

Alongside my roles in industry and academia, I’ve maintained an independent practice in software engineering, architecture and teaching. My work has spanned radar technologies, blockchain, zero-trust systems, cybersecurity for medical devices, fintech and cloud services, alongside corporate training.

Through this practice, I also wrote a freely available Python book and developed a metaprogramming engine that turned high-level blueprints into complete systems. That work became an open-source project, Auryn, and a second free book.