The LlamaCon Hackathon, a celebration of AI bringing together developers from across the globe, successfully concluded in San Francisco. The event garnered over 600 registrants, with 238 talented developers and innovators ultimately converging to build projects over an intense 24-hour period. The challenge: create a demonstrable project within the given time frame, leveraging the Llama API, Llama 4 Scout, or Llama 4 Maverick – or any combination of these cutting-edge tools.
The competition featured a substantial prize pool, with $35,000 in cash rewards distributed across first, second, and third-place finishers, as well as a special "Best Use of Llama API" award. A panel of judges from Meta and sponsoring partners meticulously evaluated the 44 project submissions.
A tremendous thank you to our partners: Groq, Crew AI, Tavus, Lambda, Nebius, and SambaNova, who provided invaluable support throughout the hackathon. Each sponsor contributed credit usage, workshops led by expert speakers, mentorship, on-site Q&A booths, judging participation, and remote assistance on Discord.
And The Winners Are …
After two rounds of judging, the top six projects were selected from the 44 submissions, ultimately resulting in the determination of first, second, and third-place winners, as well as the recipient of the “Best Use of Llama API” award.
OrgLens – First Place Winner
OrgLens has successfully created an AI-powered specialist matching system that connects you with the right professionals within an organization. By analyzing data from various resources, including Jira tasks, GitHub code and issues, internal documentation, and resumes, OrgLens creates comprehensive knowledge graphs and detailed profiles for each contributor. This empowers you to search for experts using advanced AI-powered search or even interact with a digital twin of the professional, allowing questions to be asked before reaching out. A demonstration Web application was built using React, Tailwind, and Django, leveraging the GitHub API and Llama API for data processing and storage, to showcase its capabilities. OrgLens streamlines expert matching, making it easier to find the right person for the job.
Delving deeper into the innovations of OrgLens, it is more than just an expert matching system – it’s an accelerator for internal knowledge sharing and collaboration within an enterprise. It ingeniously leverages the power of artificial intelligence to break down information silos and connect the expertise hidden across the organization. Imagine confronting a complex project and, instead of aimlessly searching through internal emails and documents, quickly finding colleagues with relevant experience and skills through OrgLens. The ability to initiate preliminary communications with their “digital twins” further enhances efficiency and problem-solving speed. The core strength of OrgLens lies in its deep data mining and analysis capabilities. It not only crawls data from platforms like Jira and GitHub but also analyzes internal documents and resumes to build a comprehensive knowledge graph. This knowledge graph includes employees’ skills, experience, contributions, and interactions in different projects. Through this knowledge graph, OrgLens can accurately identify the best expert for a particular task and recommend them to those in need the help.
Furthermore, OrgLens emphasizes user experience. It provides an intuitive and easy-to-use Web interface where users can search by keywords or use advanced filters to find the suitable expert. The integration of the “digital twin” feature further elevates the user experience, allowing users to ask preliminary questions and get quick answers to save time for both the experts and seekers. By embedding artificial intelligence into the expert matching process, OrgLens possesses the ability to revolutionize the way companies manage and utilize their internal talent resources, leading to improved collaboration, innovation, and overall performance.
The success of OrgLens stems from its solution to a widespread knowledge management problem within enterprises. Many companies face the challenge of dispersed employee skills and difficult-to-access information, leading to wasted resources and inefficiencies. By automating the expert matching process, OrgLens effectively addresses this problem and brings several benefits to the enterprise:
- Improved Productivity: Employees find help they need faster, accelerating project progress.
- Fostered Innovation: Connecting experts from different fields can spark new ideas and solutions.
- Optimized Resource Utilization: Avoids duplicate work and resource waste, improving overall efficiency.
- Enhanced Employee Engagement: Makes it easier for employees to share knowledge and experience, enhancing their engagement and belonging.
Compliance Wizards – Second Place Winner
Compliance Wizards created an AI-powered transaction analyzer for detecting fraud and alerting users based on custom risk assessment algorithms. Email notifications are sent to users, prompting them to report or confirm a transaction. Users can then engage with an AI voice assistant to report and acknowledge the transaction. Leveraging the Llama API’s multimodality, fraud assessors can upload customer information and search relevant news regarding their client to help determine if the client is involved in any noteworthy criminal activity.
Compliance Wizards has created an AI-powered transaction analyzer which is meticulously designed to identify suspicious activities and alert users through intricate risk assessment algorithms. The system functions by sending email notifications to users, prompting them to review and confirm certain transactions. From there, users can interact with an AI-driven voice assistant to either report suspicious transactions or acknowledge its legitimacy. By utilizing the Llama API’s multi-model capabilities, fraud assessors are enabled to upload customer information and search for relevant news that could help in discovering noteworthy criminal activity.
At its heart, Compliance Wizards features a very powerful AI engine that can thoroughly analyze transaction datasets and identify potential fraud patterns. This engine not only detects traditional fraud behaviors but is also capable of customizing risk assessments based on specific client risk profiles, improving the accuracy of fraud detection. Additionally, Compliance Wizards integrates the news search functions, helping the fraud assessors rapidly gather the relevant information on their clients, gathering information such as media mentions and legal records. The contextual information is crucial to assessing the client’s overall risk profile for any potential red flags.
The AI-powered voice assistant is a key element in Compliance Wizards. It offers users an efficient and convenient way to report and acknowledge transactions, especially while traveling. The voice assistant is able to answer any question about the transactions and assists on how to obey regulations.
A multi-layered approach towards security is the prime advantage of Compliance Wizards:
- Advanced Risk Assessment: Delivers a customizable risk assessment algorithm to identify fraud behaviors.
- Real-Time Transactions Analysis: Performs real-time monitoring on all transactions to promptly identify suspicious activities.
- Context-Aware: Helps the fraud assessors capture news data, assisting assessors with the assessment of client’s conditions.
- Simplified Reporting: Offers a voice assistant that simplifies the reporting and confirmation process.
Compliance Wizards is more than a tool; it’s a complete compliance solution that aids corporations with mitigating risks and adhering to regulations.
Llama CCTV Operator – Third Place Winner
A team led by Agajan Torayev built a Llama CCTV AI control room operator that can automatically identify custom surveillance video events without any model fine-tuning. The operator is able to define video events with simple language. Using Llama 4’s multimodal image understanding, the system captures and detects motion every five frames to assess these predefined events and report them to the operator.
The underlying concept is the ability in empowering surveillance systems, enabling the ability for actively identifying anomalous events rather than just passively recording. By utilizing the robust image understanding capability, Llama 4 can analyze video feeds in real-time, thus identifying a wide array of events, such as potential suspicious actions, unauthorized entry, or safety threats. The operator can define these events in simple language, not requiring any specialized skills nor knowledge in machine learning or computer vision.
Following operations: system captures and examines actions within every five frames; then utilizing Llama 4’s features to evaluating if captured actions match any of the predetermined events. If an action matches the predefined events, the system will rapidly report it to the operator with relevant information.
The key competitive advantage is shown in Llama CCTV Operator:
- No Fine-Tuning Required: Simplifies the installation and sustainment process; remove the need of model fine-tuning.
- Custom Event Detection: Offers an operator to define customized security events, serving certain security situations.
- Real-Time Analysis: Analyses video feeds in real-time, identifying suspicious actions promptly.
- Automated Reporting: Decreases the labor control requirement, automatically reported detected events to operators.
Geo-ML – Best Use of Llama API
Geologist William Davis used Llama 4 Maverick and GemPy to generate potential mining locations, topographic maps, and 3D geological models of mineral deposits. Geo-ML works by processing 400-page geological reports, consolidating the information into a structured geology domain-specific language, and then using it to generate 3D representations of the subsurface geology.
"This was my first time actually using an LLM API to extract super-long text and images from long-form geological research papers, so I used the super long context window of Llama Maverick as well as text and image multimodal functionality to extract the text and convert it into a domain-specific language to give a compressed version of everything stored in the documents," Davis said. "I spent most of my time reading geology documents. It would be great to have an LLM that could do that for me in the background."
William Davis, geologist, creatively utilizes Llama 4 Maverick and GemPy to pioneer a brand new methodology of geological modeling. Geo-ML’s purpose is extracting data from the large geology reports and turning them into understandable 3D models.
Geo-ML processes extended geological reports and consolidating the information into a domain-specific language. This language represents key and geology, which are described previously in the reports. After that, the system applies that language in order to create 3D representations of underground geography.
Davis emphasizes the long context window from Llama 4 and multi-model capabilities in manufacturing Geo-ML can analyze the whole research papers, and extract both text and images.
Geo-ML has several advantages in:
- Automated Geology Modeling: Automates geo modeling, also minimizing manual time of labor analysis.
- Extract Key Information: Helps geologists with extracting key data and potentially locating mining and minerals.
- Produce 3D Models: Helps people easily analysis the underground settings using 3D representations.
- Accelerated Geology Research: Speeding up the research with accelerated geol modeling.
Honorable Mention: Team Concierge
One finalist, named Concierge, brought their own GPUs to the competition, which helped them stand out from the crowd.
"We believed that the best aspect of Llama 4 Maverick was its sparse mixture of experts nature and open-source availability, allowing for fine-tuning," the team said. "Meta recently released an excellent fine-tuning tool, that tool being on GitHub. Using the Llama API, we compiled data from multiple sources to create QA datasets and fine-tuned the Llama 4 Maverick model. We plan to submit this to open benchmarks, because we currently lack a Llama 4 encoder, and with the 1M context window, it has the potential to be exceptional."
Concierge’s unique point is it focuses on calibrating Llama 4 Maverick model with improved performance for specifications. The team believes the open source and mixture aspects from Llama 4, making the model as the best candidate for calibration.
The team gathers from data resources in order to create the QA datasets and training the model later on. The team decides to submit the calibrate model to public benchmarks for the result assessment.
Watch Demonstrations of Finalists
You can watch the demonstrations of finalists on YouTube.
Join the Next Llama Hackathon
Developers can apply to participate in the next Llama Hackathon, which will be held from May 31 to June 1, 2025, in New York City.