Icebreaker convoys on the Northern Sea Route
When each ship sails alone and when it joins a convoy, hour by hour, as the ice changes from week to week.
Fuel distribution · trips from two depots →
Operations research engineer

Operations research engineer in Belgrade with 7+ years of building optimization software for logistics, crew scheduling and industrial planning. I take a planning problem from the first formulation to a service in production, and I build OR Lens, an open-source plugin that lets AI agents inspect optimization models.
ivan@rodn.devLinkedInGitHubCV (PDF)
When each ship sails alone and when it joins a convoy, hour by hour, as the ice changes from week to week.
A target apartment mix turned into buildable floors: facades, lift-and-stair cores, shafts, wet zones and daylight rules.
How much fuel each filling station receives and when, then which tanker drives which trip, within tank capacities and station time windows.
Sports timetables where every team meets every other twice, under capacity, break, separation and fairness rules.
A day of arrivals and departures on 275 stands: jet bridges, buses, taxiing and wide-body neighbors.
A deposit forecast for every terminal, then armored-car routes chosen day by day for three months.
Admissions, operating theaters and nurse rosters as one problem, with ten minutes per instance.
Cover the work first, then give out vacations and grant requests, in strict order of priority.
How much to buy and how much to make at each stage and site, week by week, with lead times, minimum orders, stock targets and product changeovers.
A week of production across several production centers: stages of fixed length on shared lines, then warehouses that ship to customers, against sales and stock targets.
Which plants and warehouses to open at each tier and how goods flow from suppliers to customers, at transport tariffs from a rate database. An AI agent helps the analysts interpret the results.
Air-crew rosters and retail shifts: coverage, qualifications, labor rules and preferences as hard and soft constraints.
Routes and movement schedules for a long-haul transportation network, planned together under capacity and timing constraints.
Operations of every order on shared machines, each order following its own route through the shop.
A live rental search for Belgrade: listings from portals and Telegram, map search, photo deduplication and ranking by taste.

Debug optimization models with facts, not guesses.
A plugin for Claude Code and Codex. Ask about an LP or MPS file in chat: HiGHS extracts the conflict, an independent re-solve confirms it, and the workbench opens beside the chat.
Statuses, conflicts and objective values come from HiGHS and independent checks, never from the language model.
claude plugin marketplace add ivs-rodin/or-lens-plugin claude plugin install or-lens@or-lens
Explains why an optimization model is infeasible.
A Python library for Pyomo models and LP or MPS files. It splits the model in halves, first by constraint groups and then inside them, until only the conflict is left, and says what to switch off. OR Lens will use it next.
pip install git+https://gitlab.com/tarasov.alexey/stolp.git@main