# Bohdan Fesenko > Bohdan Fesenko: engineering lead and systems engineer. Agent platforms, measured detection and tracking, reviewable AI-assisted development. Bohdan Fesenko is an engineer in Kyiv who builds systems where AI agents and people do real work together. He led engineering at a pre-MVP consumer startup, took a five-person team from no process to a shipped product, and made agent-assisted development a reviewable practice. He now builds agent platforms and measured detection-and-tracking pipelines. ## Case studies - [track-lab: from noisy detections to confirmed tracks, measured](https://fesenko.ai/work/track-lab.md): Targets too faint for any single frame, noise that looks like targets, labels that arrive late. A reference tracker and a simulator that knows the truth, with every decision measured and runnable in the page. - [Agent-assisted development as a reviewable practice](https://fesenko.ai/work/reviewable-agentic-development.md): I made AI sessions part of code review, with a rubric, named red flags and an onboarding gate. Velocity without reviewability is debt. - [An MCP surface that refuses to boot if a tool lies about itself](https://fesenko.ai/work/mcp-surface.md): 83 tools, five live resources over server-sent events, OAuth with dynamic client registration for third-party clients, and a registry validated at start-up. - [An assistant that operates this site in front of you](https://fesenko.ai/work/site-agent.md): Ask it something and it flies to the page, points at the section, runs the demo and turns its knobs, then answers with cards built from the site's own content, so it cannot invent a number. - [Few-pixel targets from a moving camera, in real time on the edge](https://fesenko.ai/work/moving-target-indicator.md): A target three pixels wide carries almost no information in space. All of it is in time, so the pipeline integrates before it decides. - [One memory for a fleet of coding agents](https://fesenko.ai/work/agent-memory.md): Several agents, several machines, one shared data layer with mounts, signal streams, addressed whispers and a secrets vault, and a rule about who may write what. - [A mobile product shipped to TestFlight in six weeks, by one engineer with agents](https://fesenko.ai/work/fara.md): A social app for a city's motorcycle community, with a Telegram bot for the chats riders already use: iOS and Android from one codebase, one verification gate shared by people and CI. ## Demos - [The pipeline](https://fesenko.ai/demos/track-lab): Seven layers of a detection and tracking system, each with a working miniature. - [Live tracker](https://fesenko.ai/demos/track-lab/live): Targets, clutter and a tracker in the browser. Sliders for SNR, threshold and confirmation policy; counters for false tracks per hour and time to confirm. - [Where simple trackers break](https://fesenko.ai/demos/track-lab/cases): A sharp turn and two targets crossing: three methods on the same data, one recorded run side by side, and the measured rates over many runs, with their noise. - [Invisible target](https://fesenko.ai/demos/track-lab/tbd): A 3 dB target no frame can show, found by integrating along motion hypotheses, priced at matched false alarms. - [Bearing-only localisation](https://fesenko.ai/demos/track-lab/fusion): Drag sensors and a target; the Cramér-Rao ellipse and the geometry that breaks a fix. - [Imperfect truth](https://fesenko.ai/demos/track-lab/gt): Which metrics still rank trackers correctly when labels are late, sparse or missing. - [Results](https://fesenko.ai/demos/track-lab/results): The numbers from the Python experiments. ## Knowledge Base - [Does INT8 quantization hurt small object detection? The damage mAP does not show](https://fesenko.ai/knowledge-base/int8-tiny-targets.md): Quantizing a detector to INT8 costs small objects 1.4 to 3 times more than the overall metric, and below 12 pixels the published loss reaches a third of the accuracy; what is known, the fixes, and what nobody has measured yet. RSS feed: https://fesenko.ai/knowledge-base/feed.xml ## Pages - [Home](https://fesenko.ai/) - [Markdown](https://fesenko.ai/index.md) - [Work](https://fesenko.ai/work) - [Markdown](https://fesenko.ai/work.md) - [Demos](https://fesenko.ai/demos) - [Markdown](https://fesenko.ai/demos.md) - [Challenges](https://fesenko.ai/challenges) - [Markdown](https://fesenko.ai/challenges.md) - [Knowledge Base](https://fesenko.ai/knowledge-base) - [Markdown](https://fesenko.ai/knowledge-base.md) - [Library](https://fesenko.ai/library) - [Markdown](https://fesenko.ai/library.md) - [CV](https://fesenko.ai/cv) - [Markdown](https://fesenko.ai/cv.md) ## Languages Every page, the demos under /demos/track-lab included, also exists in Ukrainian under /uk (for example https://fesenko.ai/uk/work); every page except the interactive demos has a Markdown twin, under /uk as well (https://fesenko.ai/uk/index.md). ## Contact - [bohdan@fesenko.ai](mailto:bohdan@fesenko.ai) - [GitHub](https://github.com/bohdan-fesenko) - [LinkedIn](https://www.linkedin.com/in/bohdan-fesenko/) ## For agents The site also exposes its interactive actions to browser agents through WebMCP. These actions are available through document.modelContext when the browser supports it.