Projects building at the edge of AI and sustainability.

When everything shuts down, our app keeps people connected, informed, and prepared.

When a blackout happens, every second matters. Our app provides a centralized emergency communication system designed to keep people informed, connected, and calm — even during network disruption. Once an authorized coordinator, such as a mayor or crisis manager, activates emergency mode, all users receive an urgent full-screen alert explaining the situation and limiting unnecessary phone activity to prevent network overload. Inside the app, users can instantly notify selected contacts that they are safe and access real-time updates from local authorities. The app also includes an AI-powered crisis assistant that helps people handle common emergency situations without overwhelming emergency services. The assistant provides practical guidance, answers urgent questions, and evaluates when professional help, such as an ambulance, is truly needed. Meanwhile, coordinators can continue sending live updates and instructions to the community throughout the crisis. Our goal is simple: reduce panic, improve coordination, and help people stay prepared when traditional communication systems fail.

GridPulse turns 8,760 hourly power grid snapshots into an interactive, AI-powered decision support dashboard - answering what's happening, what's forecast, and what if a line fails

GridPulse is a real-time decision support dashboard for the ČEPS transmission grid, built for GreenHack 2026. It uses Streamlit + Pandapower + Plotly to visualize 8,760 hourly snapshots of the IEEE 118-bus system across 5 pages (Overview, Grid Map, Forecast, Simulations, Drill-Down), with an AI chatbot (DeepSeek LLM + rule-based fallback) that answers natural-language questions, runs SQL queries, and simulates what-if scenarios like line outages or load scaling — all wrapped in a neon UI.

Agentic system to help grid operators make better-informed decisions and to help keep the grid smooth

Grid operators have to make informed decisions about the grid, where to route the energy, what generators to start stop etc. Navigating the wast amount of data is complicated (think hundreds of generators, loads and branches). They obviously already use a bunch of classical tools to help. Could agents help them navigate this? Typical question to imagine: A line is nearly overloaded, why is this happening? So what does this mean? We have a map with a chat window (you've seen this before) with the grid information. What can I do with it? Lets start with basics and move on to better stuff: >> What is the grid load? >> What's the branch/hub with highest load? - Works, but basic >> What's most vulnerable branch? - runs N-1 contingency and fetches details about the branch >> Why? What if it overloads? - looks at the topology and tries to analyse (e.g this branch has no parallel connections) - Getting intereseting, these use and chain tools >> Did today run to plan? Which metric had the biggest surprise? - Looks at forecasts, analyses deviations and presents the data >> Are any lines loaded above their normal range for this time of day?

Convince your neighbors to come to terms

There are many programs, materials, and information about green renovation and why it's important but the biggest challenge is to convince your neighbors to vote for renovation. Our project is the step to help champions persuade conservative minds

A hands-on platform where employees try AI for real, review each other's results, and leave the room less afraid of it than when they walked in.

Most people who are hesitant about AI aren't hesitant because it's hard. They're hesitant because they've never had a safe, structured place to try it with someone watching their back. This platform gives them that place. Participants pick up a practical task — summarise a meeting, draft an email, fact-check a report — and use whatever AI tool they want. They submit their attempt. A peer reviews it against a shared rubric and writes back: what worked, what didn't, what to try next. Nobody presents to the room. Nobody's output goes on a slide. It's just two colleagues, a task they both tried, and an honest conversation about what the machine actually did. Facilitators set the tasks and rubrics. The rest runs itself.

An AI grid co-pilot that uses real power-flow simulation to reveal overload risks, carbon impact, and mitigation paths for a greener Czech transmission grid.

Grid Pulse is an interactive transmission-grid stress-testing platform built for the GreenHack 2026 ČEPS challenge. It lets users explore an illustrative Czech/Central-European high-voltage grid, select grid conditions, trigger branch failures or Dunkelflaute-style stress scenarios, and see how line loadings, overloads, generation mix, and carbon intensity change. A physics solver, not the LLM, computes the grid state using pandapower DC power flow; the AI co-pilot translates those results into clear operator-style explanations and can answer questions about overloads, reserves, dispatch, and cascade risk. The goal is to make complex grid resilience tradeoffs visible: where the grid is vulnerable, why it is vulnerable, and which interventions such as storage, redispatch, or interconnection support could reduce risk.

Demo link not provided

Grid Pulse gives grid operators real-time situational awareness through an AI map and chatbot, plus a what-if mode that runs a real power flow on their edits — so they see a line

Grid Pulse is an AI-powered, real-time decision-support dashboard for electricity transmission grid operators, built for GreenHack 2026 / ČEPS. It turns the firehose of grid telemetry into something an operator can see and question at a glance. An interactive deck.gl map of the IEEE 118-bus network (118 substations, 186 lines across three sub-regions) shows the live system pulse, against the exact same grid state. Its standout feature is Simulation Mode: a split-screen REAL | SIMULATED view where the operator runs what-if edits — reconnect a bus, trip a tie-line, island a whole sub-region, or shift load and generation — and the right pane re-solves a real AC power flow (pandapower) to show how the grid re-routes, before any action is taken. ML overlays flag congestion and blackout risk ahead of time.

Demo link not provided

Your first step to collective renovation

renovuj.me helps changemakers to find conviction and support to push their ideas and find a way how to make a shared residential spaces a better place.It helps by breaking the human stalemate that blocks them at owner meetings, unlocking a massive, frozen pipeline of residential financing. Currently, billions of crowns in commercial green loans and state subsidies sit unspent simply because housing association (SVJ) members cannot reach a consensus to approve projects. By equipping a single motivated building resident with clear, per-flat financial math and persuasive pitch materials, our platform turns complex, easily rejected proposals into undeniable, vote-winning business cases. This creates a high-conversion lead generation engine for partner banks and property managers while driving profound real-world impact: moving stuck renovation projects from inertia to action to permanently slash the housing sector's carbon footprint, upgrade aging urban infrastructure, and protect everyday citizens from volatile energy costs.

Grid Pulse turns grid data into a live dispatcher map with KPIs and N-1 risk flags; the AI proposes a data-backed fix, the human approves, and shift summaries ease handovers.

Grid dispatchers work long shifts under heavy data load. Grid Pulse gives them three things that make the job easier. First, a shift handover summary: at the end of a shift the AI writes a clear, data-grounded report of what happened, so passing the grid to the next dispatcher is effortless. Second, alerts backed by real data: built on ČEPS's IEEE-118 dataset and real pandapower load-flows, the live map auto-flags risks — including N-1 contingencies, where losing a single line would overload another — with the actual numbers behind every alert. Third, an AI assistant: whenever a dispatcher is unsure, they can ask in plain language and get answers drawn from the real grid state, never fabricated. Throughout, a human stays in control — the AI proposes a data-backed solution and the dispatcher approves it. The result: the grid made visible, understandable, and safe to hand over.

Demo link not provided

A functional web application featuring AI emergency triage, offline survival guides, and the architecture for a P2P mesh network.

Prague Now is a city resilience web application designed to help citizens stay informed and prepared during blackouts and urban emergencies. It combines offline-ready survival guides, interactive critical infrastructure maps, volunteer coordination, official updates, and AI-assisted emergency advice in one municipal super-app. The project is built as a TypeScript monorepo using pnpm Workspaces. The frontend uses Next.js 14, React 18, Leaflet maps, Lucide icons, and custom CSS theming with dark/light modes. The backend is based on Node.js, Express, Prisma, and PostgreSQL for managing users, markers, guides, and community data. Shared Zod schemas and TypeScript models keep validation consistent across the frontend and backend. For offline communication, Prague Now plans a peer-to-peer mesh layer using Bluetooth Low Energy for discovery and signaling, plus Wi-Fi Direct or local Wi-Fi for message exchange where available. Messages would use end-to-end encryption, signatures, and store-and-forward routing. AI support with Claude Haiku through a protected server-side API route, using emergency workflows, automations, input filtering, and a master prompt to prevent misuse.

We enable data-driven decisions on whether an EV is a suitable choice given the driver's weekly routine.

Our app helps potential buyers determine if an EV fits their lifestyle by analyzing their weekly driving routine. It recommends the best charging spots and times, minimizing the number of charging sessions, detours, and costs.

A database based platform of volunteers ready to help in a crisis

Ready2Help is a crisis-response platform that bridges the gap between willing volunteers and emergency authorities. It provides a centralized, interactive database where citizens can register their locations, skills, and available equipment (like cars or generators). When a crisis hits, authorities can instantly filter this data, visualize volunteer locations on a map, and export lists for field operations, ensuring the right help gets to the right place as quickly as possible.

Grid Pulse is an AI-powered platform that visualizes the real-time state of the power grid, helping operators detect risks, understand trends, and make faster decisions.

Grid Pulse is an AI-powered platform designed to help power grid operators better understand, monitor, and anticipate changes across the power grid. Using historical grid data, we trained predictive model to identify patterns and forecast potential future developments in grid behavior, giving operators valuable insight before issues occur. Our solution combines real-time grid flow monitoring with an intuitive visual interface that displays the current state and health of the network. Operators can quickly identify critical areas, track energy flows, and gain a clear overview of the system without having to interpret large volumes of raw data. To make information even more accessible, Grid Pulse includes a locally deployed AI assistant powered by Ollama and qwen2.5:7b. Operators can interact with the system using natural language, ask questions about grid conditions, investigate anomalies, and receive contextual explanations directly from the platform. By combining predictive analytics, visual situational awareness, and conversational AI, Grid Pulse transforms complex energy data into actionable insights.

Demo link not provided

BlacOut keeps Prague connected when everything else goes dark.

BlackOut is a mobile application designed for large-scale power outages. It provides critical information, offline maps, emergency contacts, and communication through a Bluetooth mesh network, helping people stay connected, informed, and prepared even when internet and mobile networks are unavailable.

Volta is a functional engineering platform built in Go that connects real pandapower network simulation with self-hosted LLMs in Docker, allowing dispatchers to predict over

Our project, Volta, revolutionizes the dispatch control of energy grids. We developed a fully functional engineering platform that connects real data with the pandapower physics simulation engine and modern artificial intelligence. Our main goal is to eliminate information overload for dispatchers. The system parses network JSON states in real-time, visualizes them on an interactive map, and constantly analyzes day-ahead weather and consumption forecasts using Gemma and Qwen LLM models running in Docker containers. If the AI detects an upcoming line overload, it instantly alerts the dispatcher and suggests specific steps for grid reconfiguration.

Reno translates complex building data and state subsidies into clear dashboards and tailored arguments.

Upgrading apartment buildings is a logistical and social nightmare, and Homeowner Association (SVJ) meetings often fail because chairpersons lack clear data to persuade residents. Reno solves this as an automated co-pilot: the user simply inputs an address with basic energy information, and the app automatically fetches public cadastre data to generate a customized renovation roadmap. It calculates the total investment, pairs it with state subsidies, and visualizes the exact financial impact down to a single apartment unit, while also generating tailored counter-arguments for specific neighbor personas and exporting personalized presentation decks.

AI cost-control and routing layer for LLM applications - cut LLM spend without sacrificing quality.

Argus is a control room for a fleet of AI agents. The frontend renders each agent as a character in a themed room; the Argus backend is a token-governance engine that routes, caches, dedups and watches every LLM call — flagging agents that waste tokens, loop, or blow their budget.

When the grid goes down, Nexus keeps you connected - turning nearby phones into a mesh network so communication survives the blackout.

The Prague blackout of July 2025: one conductor failure, 90 minutes, and a city went silent. Mobile networks saturated, hospitals on generators, no real coordination. Nexus is built for day 2 - and day 5. A mobile app with two modes. Online, it works like any other tool: messaging, cloud AI assistant, government alert feeds, household preparedness checklists. When the grid drops, it switches to a WiFi Direct mesh. Each phone becomes a node. Messages hop device-to-device across up to 100m without touching a router. Routing decisions are made on-device by a trained agent that adapts to dying batteries, dropped nodes, and shifting topology. A compact language model runs as an offline local AI chatbot - no internet needed for CPR instructions. Five screens: SOS, communicator, preparedness checklist, AI chatbot, crisis map. Works on the phones people already own.

Demo link not provided

A solar-powered emergency mesh network that keeps critical infrastructure and first responders connected, monitored, and remotely controllable during outages.

Signal in Dark is an emergency communication network designed to keep critical services operational during prolonged blackouts. The system consists of solar-powered mesh nodes deployed at strategic locations and connected through the Reticulum networking stack. Depending on distance and infrastructure availability, nodes communicate using LoRa (essentially a digital “walkie-talkie” capable of sending small messages over long distances) or Wi-Fi HaLow for faster local connections. The network is intended for hospitals, emergency operations centers, police and fire stations, power plants, water treatment facilities, and other critical infrastructure. During large-scale outages, when mobile networks become overloaded or fail completely, the mesh provides an independent communication layer for exchanging messages, sharing status information, and coordinating response efforts. Beyond communication, nodes can be integrated directly into infrastructure systems to enable remote monitoring and backup control of equipment. If a facility loses its primary network connection, operators can still receive alerts, monitor system status, and control essential components through the mesh network.

Zeus brings revolutionary autonomous AI technology to power grid management.

Zeus is a power grid management platform built upon the base GridPulse implementation spec. We implement all of the basics from the spec sheet, but we have an ace up our sleeve— Our AI dispatcher named Athena. This revolutionary, multi-modal, multi-agent AI is capable of keeping the dataset simulation alive, autonomously for up to 10 hours with 100% accuracy. 20 hours with 75% accuracy and so on... The best part? This accuracy is achieved on AI models that are literally available for free. And also, we built a power grid simulation, which factors in every piece of data we could get our hands on — creating a really comprehensive testing space for the model.

From the first forecast to a signed renovation. A guide that leads an apartment building to a decision.

One app that first calculates the renovation, then explains it clearly, and finally guides the whole homeowners’ association toward a binding decision — step by step.

Keeping the lights on: our AI helps power plants prevent blackouts through real-time grid optimization, with humans retaining full control at every step.

We developed an AI-assisted dashboard that provides a real-time overview of power plants and grid load through an interactive map. The system continuously monitors network conditions, detects potential overload risks, and recommends optimal actions such as power rerouting to maintain grid stability and prevent blackouts. All critical decisions remain under human control, with AI serving as a decision-support tool. The platform also includes advanced security features, including facial recognition and employee card verification for secure access management. In addition, users can generate comprehensive PDF reports containing historical data, performance metrics, and operational statistics for analysis and compliance purposes.

Demo link not provided

PROMPTA helps organizations reduce AI waste and improve productivity by guiding employees to use the right AI tools efficiently.

PROMPTA is an AI adoption platform that helps companies use AI tools (ChatGPT, Copilot, Gemini, etc.) more efficiently and sustainably. Instead of random and repetitive AI usage, PROMPTA introduces structured learning paths, AI tool recommendations, and shared organizational knowledge. This reduces duplicated work, unnecessary AI queries, and inefficient tool usage, leading to lower computational waste and higher productivity. The platform also tracks AI usage efficiency and helps companies optimize their AI footprint.

Demo link not provided