Notebook with my (Duarte) work at the Hack N Chill for the Spot Pricing challenge. Stream of thought data analysis treatment and modelling as I re-learned some ML and tried learning some more
This project lets the user model their expected sales outcomes through convolutions of probability distributions and quantities.
Kuluko lets users generate full, custom audiobooks, from as little as an idea for the book. We use tons of different models to write a book, turn it into audio, create covers etc. Right now we are about to launch the next major update which introduces "audio dramas", which differ from normal audiobooks by having multiple speakers, sound effects and music. For this update we are focussing heavily on finetuning open source TTS models to provide payable, custom, high quality voices for the audiobooks.
Proyecto enfocado en la creación de un Agente en IA para la Defensa y Seguridad del país. Este Agente IA tendrá la siguientes funciones: - Aplicar una estrategia de acción en seguridad y defensa del país ( pandemia, ataques terroristas, guerras, ciberataques. amenazas) basada en toma de decisiones. - Formación y Creación de grupos de trabajo nacionales e internaciales cuya misión es trasladar el mensaje indicado por el Agente IA a los organismos gubernamentales que competan de los distintos países afectados. - Establecer reglas para la Automatización de tareas Entre otras funciones
KHWARIZMI: From Thought to Vision. Our system transforms your natural language prompts into comprehensive reports with data visualizations, SQL queries, and textual explanations - turning your analytical thoughts into visual insights.
DataCrunch
A React Native app used to provide content creators with thte tools dthey need to become viral and engaging, with many tools tailored to creators such as a news feed that includes many things for creators to make videos about, there is also a personal AI clickbait finder, as well as a content analyzation tool that returns feedback and a engagement score. Teammate: Oliver Viitaniemi https://youtu.be/eWJedjwgpo4?si=4Mbi8F7bySETxcPZ
Ducker is an AI Data Analyst solution for people who are lazy to write their own sql queries, also it is a solution to the hackathon challenge by SmartBI. It uses react, next.js, shadcn, tailwind CSS, python, fastAPI, duckdb, openai and much more
H-AI, the Human Hands-Free Helper, is an AI-powered assistant that captures and analyzes everything technicians do in the field – through video, voice, and text. It transforms rugged, hands-on work environments into streamlined, error-free reporting processes. By eliminating manual data entry and leveraging AI to auto-generate post-operation logs, we ensure technicians focus on what matters most: safe, high-quality repairs. The result? Happy, hassle-free, and high-quality field service experiences – every time. Teammates: Tianran Li, Junyuan Fang, Rosanna Hu, Haoyu Wei and Yibo Zhang. Demo link: https://youtu.be/VBIQ4MedSxo
DataCrunch
DisPatcher supports maintenance workers from start to finish. It provides job details and checklists upfront, answers questions on site using a vector database of embedded manuals, and simplifies reporting by auto-generating post-job summaries from audio transcriptions and logged data. Teammates : Matús, Nazaal, Shahida, Robert, Nahl Video link: https://www.youtube.com/watch?v=gfjzBqDtbJ0
DataCrunch
ExplainDB combines natural language processing, database querying, and data visualization in the following automated workflow: ######################################################Generate SQL queries based on user questions*** Execute the queries and clean the result data*** Use the GPT-4.1 model to refine the query result and generate a natural language answer*** Automatically generate additional data exploration goals relevant to the user's question*** Use LIDA to generate visualizations and explanations*** Display a main chart and multiple recommended charts, each accompanied by a natural language explanation to help users understand the visualized data*** Support multi-turn conversations and session history to allow follow-up questions and reviews*** ###################################################### This platform provides an intuitive, all-in-one data exploration experience for users who may not be familiar with SQL or data visualization tools. ###################################################### Built by HackAvengers: Gong Qiao & Shivang Gupta
AaltoAI hackathon project for an data analysis agent
Link to the video: https://youtu.be/3i2aok9xppA. Team members: Dat Doan, Haitham Al-Shami, Hari Prasanth S.M., Muhammad Ahmad, Rohail Malik. The project is an app for Microsoft's challenge, focusing on easing elevator technician work. We use multiagents in langGraph with a neo4j data base. We included the ability to transcribe speech-to-text and a camera. The camera uses openCV. The whole project is built in a docker container and can be ran quite easily by allowing it to build.
The project presents AI platform that can understand natural language to generate SQL queries and charts. The project is a result of a tedious research of available solutions that serve this purpose and is based on WrenAI platform. Key features include: - Text-to-SQL - Text-to-Charts - AI-generated concise summary and explanation of results - User-friendly UI that serves the use cases defined by Norrin Team members: - Mehdi Moshtaghi - Ruslan Potekhin Video: https://www.youtube.com/watch?v=xOMneGifdaA
Filling out reports in the field during a service operation can be time consuming and impractical. To solve this problem, we developed a program with which you can fill these out using voice or text commands. The user can also ask for advice from the AI during the operation. Short demo video available through project link. Project made by: Pessi Fabritius, Justus Lehtola and Rasmus Mäkinen
DataCrunch.png)
This project addresses the challenge of identifying and consolidating innovation disclosures from VTT's collaboration partnerships. Team name: Agent Warrior Team members: Yang Jun, Gao Hao, Chen Yuxin, Li Lijie, Li Changrong
DataCrunch
We created a MCP (Model Context Protocol) based service that uses OpenAI and LangChain agent framework to query interesting data for the user. The agent is able to call tools and resources to figure out what response the user wants. Then if the user requests or the AI decides to make a chart we visualize it using D3.js otherwise it is simply presented as text in our frontend. We are hosting the service publicly so anyone can try it out (while credits last) at vectorquerycorns.org. ############################################### Here are some cool points of our solution: -It understands natural language -It can generate complex SQL queries based on the users request -It can transform, join and filter the tables of the database to produce accurate responses -It can perform complex deduction based on the data in the database and give suggestions to the user if what they are asking is unclear. -It can create pie and bar charts when applicable -It provides clear answers and explanations to the questions -It was designed with modularity in mind so more tools are easily added and more databases and other sources of information can easily be added without much configuration -We host it ourselves, but it can be easily run locally with npm commands or docker ################################################### Our Tech stack: Backend: Python, FastAPI, SQL, MCP Hosting: CloudFlare, DataCrunch, Traefik, Let’s Encrypt, Docker Frontend: Next.js, React, tailwind Visualization: D3.js LLM: OpenAI Agent Framework: LangChain #################################################### Example complex queries 1. Give me a diagram of the 5 most traded products during the first quarter of 1997 and tell how much money was used to trade each product. Give me the amount of money in your answer. 2. Could you make a pie chart that shows the percentage of total revenue generated per product category and give me the percentages in your answer. 3. Which 5 customers had the largest drop in spending between the first half of 1997 compared to the second half of 1997. Present it as a bar chart. (ps. database only has data from 1996 to 1998) ##################################################### Here is the team behind this project: Samuel Schmidt, Juha Ylikoski, Vesa Haaparanta
DataCrunchAuthors: Nikita Masaliuk, Evgenii Korsukov, Maxim Afteniy Main description: Jypiter lab ai analyst that gets a request sentence and converts it into a complete analysis. Run the code on your laptop and interact with the framework via comand line or telegram. Telegram feature: Great ideas about work often strike when you're off the clock. With our bot, you can quickly pull out your phone and ask any question about your data — no time lost. Voice message or just text - up to you. Use tg bot: @telewhisp_bot
DataCrunchLLM based SQL database searcher and visualizer. For Norrin challenge. Team members: Mikael Gustafsson Henry Kivijärvi Maj Nyholm Josetta Rautapää Jade Reinilä
AI-powered pipeline that transforms VTT's fragmented innovation data into a clean, canonical knowledge graph using FAISS semantic search, GPT-4.1-mini, and intelligent graph curation. Team:Artemii Ustiukhin, Ekaterina Ustiukhina, Carl Kugblenu, Mikhail Shavliuk, Daria Kriukova
Defect-Vision is an AI-powered image analysis tool designed to automatically detect and classify defects in industrial products. By utilizing deep learning techniques and anomaly detection algorithms, the system helps ensure quality control by identifying imperfections such as cracks, scratches, or missing components. Key features include real-time image processing, anomaly scoring, and a user-friendly interface for reviewing inspection results. The project aims to improve manufacturing efficiency, reduce manual inspection efforts, and enhance product reliability.
DataCrunch%20(1)%20(1).png)
GodEye is an intelligent multi-channel assistant that combines natural language understanding, data analytics, and automation to serve users via Telegram, Slack, and Web App. 🧠 Ask in plain English — get real insights 🔍 Auto-generated SQL & visualizations 📊 Interactive charts (web) & image exports (Telegram) 🧾 Insightful summaries 🔌 Ready for databases, APIs, and customer support flows Teammates: Ivan Semeniuk Zachary Burda Mateusz Sopyla
Team members: Bhabishya Gurung (me), Sakhi Hashmat Khalil, Kiran Thapalia Overview This project addresses the challenge of identifying and aggregating duplicate innovations described by different organizations. It was developed as a hackathon submission for VTT, focusing on semantic AI and large language models (LLMs) to resolve ambiguity and unify innovation records. Approach Data Integration: Merged structured innovation relationship data from company websites and VTT domain pages. Feature Extraction: For each innovation, extracted textual descriptions, full source documents, organization names, and source URLs. Semantic Similarity: Used AI-based semantic similarity (likely leveraging embeddings and LLMs) to detect potential duplicates by comparing innovation descriptions. Clustering & Aggregation: Grouped similar innovations into clusters and generated unified summaries for each cluster, ensuring source and contributor information is preserved[1]. Technologies Used Python (Jupyter Notebook) Semantic AI (embeddings, LLMs) Data processing with pandas
Decks&Data is a generative slide deck tool powered by a multi-agent system that transforms structured data into customized, audience-specific presentations. Consultants, founders and executives present insights to various audiences like employees, shareholders, investors and the media. Even while working with familiar data, time is lost finding the right numbers and figuring out how to present them, which shifts the focus away from what actually matters: the message. Built on an ADK-powered architecture, the system uses a complex orchestration workflow to coordinate four specialized agents: an interpreter, a deck architect, a data analyst and an art director. Through A2A communication, the agents interpret natural language prompts, query multi-table datasets, analyze results, and generate slides with tailored charts and text. The result is clear, persuasive storytelling adapted to the goals, values, and knowledge level of each audience. The system demonstrates agentic reasoning, natural language understanding, dynamic A2A coordination, scalable orchestration, and audience-aware insight generation. Team Complonkers: Aino Hukkanen, Mael Chauvet, Anton Shumilin, Mikko Suhonen, Ilia Zalesskii.
Pranish Kumar Munnangi, Abdelaziz Ibrahim, Uzay Yildiztaskan, Vik Kopplinger
Names: Cosmo Paul, Ahmed Tolbh, Duc Le, Udayanto Dwi Atmojo, Yun Wei. Service technicians might have trouble recalling critical information for their job at hand (elevator repair, electrician, etc.). We created a mobile app that supports technicians by interpreting voice and video input and delivers step by step instructions based on custom company information for the problem at hand. Saves time, increases productivity, and can be used for data gathering on the job (video, voice, logs). Video link: https://drive.google.com/file/d/1ap5kJhfan1SUc866bwmwIsuDbaar2-qf/view?usp=sharing
Quantum optimization algorithm with AI for Kone Elevators for fast repair. Use the Powerpoint presentation in the GitHub (Tech presentation final.pdf) for listening the video provided in Project Link.
LLM Aggregators: Samu Toljamo, Olli Glorioso, Viljami Hakkarainen and David Ramos Our Three-Step Approach: Step 1: Group Similar Innovations using embeddings Generate semantic embeddings from innovation descriptions and titles Use similarity thresholds to identify potential duplicate clusters Scale analysis across thousands of innovation records Step 2: Validate Groups with LLM Azure OpenAI reviews each cluster for false positives Removes incorrectly grouped innovations with detailed reasoning Ensures high precision while maintaining recall Step 3: Aggregate Results with LLM LLM combines information from multiple sources about the same innovation Creates unified innovation profiles preserving all source details Maintains full traceability while consolidating descriptions You will find some visualizations from the project link and in the repo the most interesting file is the main.ipynb file. Video: https://drive.google.com/drive/folders/1ZdlPXga2n17u7B9Z9KeLhf-8IOcioF5p?usp=sharing
Please watch the video for a quick overview. Video Link: https://youtu.be/TMGar__tYC4 What is FieldVerse Your extra hands and eyes on the field, add on to the human experience. FieldVerse takes the simple tools you have on you like smart watch, protective glasses, software and elevates these things to make any job faster and safer. There are three key elements. Office Frontend: More than a simple worker's hub -Provide detailed instructions before the job -Provides all needed equipment for the job and provides their location and usability status -Map of the venue for locating the problem easily, as well as all relevant elements like routers, power outlet and Ethernet port -A detailed library of the past, present and future problems and how to solve them. Their symptoms and steps. -AI software creates automatic notes while on the job, the worker can add his own using speech to text. -Guarantees that all changes to the area are recorded and noted. Smart Watch: Adding to its handiness -Makes notes of sudden movements to ensure delicate parts aren't damaged. -Keeps track of posture, heartrate and potential falls of the worker, ensuring their safety and well-being. -Using Doublepoint's gesture recognition software, the worker can pull up information and notes on his goggles if he has his hands entirely tied. XR Glasses: Simple protection becomes your second pair of eyes -Guides the field engineer with troubleshooting and repairing. -Films and takes pictures of the process for future recollection. -Collects customer service data for improvement. -Collects verification of each safety step being completed. -Keeps track of rules and regulations as well as reminds of the most common mistakes to ensure compliance. -Reminds of the GDPR rules while recording and collecting data -Shows 3D scans of the inside of routers and guides hand movement
DataCrunch
This project aims to design an AI assistant that supports electricians during the maintenance of remote electrical substations. The system is built to streamline the entire repair workflow with the following core functionalities: 🔧 Before Maintenance: Intelligent Preparation Upon receiving a fault alert, the AI provides technicians with a comprehensive pre-maintenance briefing, including: Basic information about the substation Device specifications and status Local weather conditions to help plan repair schedules Required tools and recommended replacement parts 🤖 During Maintenance: Interactive AI Support The electrician can ask the AI questions such as: “Is A broken?” → “Yes, based on abnormal readings in relevant indicators...” “Should I check B?” → “Yes, according to historical cases and field conditions...” “Do I need to replace C?” → “C has exceeded its expected lifespan. Replacement is recommended.” The AI responds based on real-time data, equipment age, fault patterns, and a connected knowledge base. 📄 After Maintenance: One-Click Report Generation After the task is completed, the system generates a structured maintenance log, incorporating: Key repair steps Parts used AI Q&A logs Environmental context (e.g., weather, conditions) Key Highlights 💡 Utilizes a large language model (LLM) in combination with structured databases to answer technician queries in natural language 🔌 Supports integration with various databases and is designed for high portability 📋 Standardizes the maintenance process and reduces errors ⏱️ Helps workers prepare efficiently before deployment, saving time and increasing on-site effectiveness
Our project is an AI-powered field service assistant designed to support elevator technicians across the full maintenance cycle. It integrates ticket data, service records, manuals, and telemetry signals to provide real-time troubleshooting recommendations, detects anomalous sensor data, offers voice-guided repair support, and generates structured incident reports. Key features include a multi-agent architecture and knowledge-grounded LLM responses. It enables end-to-end workflow automation — from ticket pick-up to final report generation to significantly improving efficiency, consistency, and technician experience on-site. Group member: Ruikang Tao, Ruifeng Nie, Zimeng Zhang, Yunxuan Liu, Tao Yu
