Presenters
Source
Agentic Evolution Hackathon: Where Ideas Meet Innovation! 🚀
The Agentic Evolution Hackathon, a collaboration between MongoDB and Cerebro Valley, brought together over 200 ambitious hackers in London, culminating in more than 70 groundbreaking projects. After intense deliberation, the top three teams presented their innovative solutions, vying for substantial cash prizes and a month’s residency at the London Founder House. Let’s dive into the brilliance showcased by these teams! ✨
The Challenge: Turning Complex Problems into Elegant Solutions 💡
This hackathon wasn’t just about coding; it was about tackling real-world challenges head-on. The participating teams leveraged cutting-edge technologies to build agentic platforms that promise to revolutionize various industries.
1. PlanPass AI: Streamlining Home Planning with Intelligence 🏡
The Problem: The UK sees one in five planning applications rejected annually, costing homeowners an estimated £900 million in wasted fees. This often happens due to obscure rules buried within lengthy council documents, leading to significant frustration and delays.
The Solution: PlanPass AI offers a seamless solution. Users input their postcode, property type, and desired build. The platform then checks local council rules and generates an interactive, compliant, cost-effective, and environmentally friendly 3D home design.
Technology Stack:
- LangGraph: Used as a multi-agent workflow, with each node representing an agent.
- MongoDB: Acts as a state store for agents, enabling workflow resumption and human-in-the-loop feedback.
- MongoDB Atlas Vector Search: Powers the compliance tab, citing exact document and page numbers from the knowledge base.
- OJI Service: Utilized for knowledge base extraction and storage.
Key Features:
- Interactive 3D home design generation in under a few minutes.
- Detailed compliance checks, with all rules passed and cited.
- Environmental impact and EPC band summaries.
- DXF and PDF files ready for council submission.
Judge’s Questions & Insights:
- Data Availability: The team acknowledged that their system relies on scraping PDFs from councils and using AI to embed this data into MongoDB. They refresh their knowledge graph weekly to ensure updated compliance rules.
- Data Gaps: If council documentation is missing or a rule isn’t in their knowledge base, the system cannot provide a definitive answer. The focus is on ensuring the policy knowledge base is comprehensive and updated regularly, ideally with expert input.
- Retrieval and Retries: The agentic workflow takes approximately three minutes to run. If a retrieval fails, it retries several times. If no answer can be provided, the system indicates that.
2. Runway Ops: Agentic Cash Management for SMEs 💰
The Problem: Late payments cost the UK economy £11 billion annually, forcing 38% of businesses to shut down. Spreadsheets can show numbers, but they can’t provide actionable insights.
The Solution: Runway Ops is an agentic cash management platform that transforms insights into action. It proactively analyzes customer and supplier behavior to optimize payment flows and prevent cash flow crises.
Technology Stack:
- MongoDB: Stores historical data and triggers new events.
- LangChain: Orchestrates a series of specialized agents.
- 11 Labs: Powers AI-driven voice calls for supplier engagement.
- Fireblocks AI: Used for generating embeddings.
Key Features:
- Deterministic calculations and forecast paths before user interaction.
- Agentic layer that pulls past behavior, identifies patterns, and coordinates actions like follow-ups and payment decisions.
- Proactive background monitoring that triggers new events based on incoming context (e.g., customer emails, bank payments).
- Semantic comparison using MongoDB Atlas Vector Search for analyzing customer interactions and payment history.
- Automated drafting of emails and suggestions for proactive actions like phone calls.
Judge’s Questions & Insights:
- Target Audience: The platform is designed for Small and Medium Enterprises (SMEs) to automate and personalize the process of chasing payments and managing supplier relationships, rather than replacing finance professionals entirely.
- Security: Conversations are only persisted for auditing purposes for about a week and then removed. Data is stored in separate MongoDB Atlas instances per SME company to ensure data isolation.
- Integrations: Future integrations include Open Banking APIs for bank payments and read-only APIs for customer communications (e.g., Gmail). Hosting will leverage AWS for horizontal and vertical scaling.
3. Volt Control: Real-Time Grid Management with AI ⚡
The Problem: In densely populated areas like London, managing energy demand from data centers, EV charging, and offices in real-time is critical. Grid operators have mere seconds to decide, risking blackouts if demand exceeds supply. In 2022, electricity costs surged dramatically to ensure parts of Southeast London had power.
The Solution: Volt Control provides a live dashboard for grid operators, consolidating data from multiple sources like demand, frequency, and weather. It uses AI to predict potential issues and recommend or automatically implement solutions.
Technology Stack:
- MongoDB Atlas: Stores time-series data from live grid and weather feeds.
- AWS: Provides the infrastructure for running agents.
- Amazon Bedrock: Powers AI functionalities.
- Voyage AI: Used for generating embeddings and powering historical analysis.
- Novelight: Utilized for ultra-fast agent responses.
Key Features:
- Ingests live grid and weather feeds every 30 seconds from sources like Elexicon and National Grid APIs.
- Provides a unified view of relevant data, explaining why an issue is occurring.
- Uses Voyage AI and MongoDB Atlas Vector Search to analyze historical situations and identify successful past responses.
- Identifies flexible assets (EV chargers, office buildings) and can suggest actions like slowing down chargers or adjusting building temperatures.
- Offers automatic execution for low-risk events and an approval queue for human oversight.
- Simulates dispatching commands and continuously monitors their effectiveness, logging all actions in MongoDB.
Judge’s Questions & Insights:
- Target Audience: Primarily national grid operators who currently piece together data from disparate sources.
- Partnerships: Potential for partnerships with government entities to access real asset APIs.
- Agentic Component: AWS runs multiple agents in the background (monitoring, historical, auditing), working together. Novelight is used for cost-effective, fast responses. The team even hit their credit limit during the hackathon, highlighting the intensive use of AI services.
The Verdict: Celebrating Innovation and Excellence 🏆
The Agentic Evolution Hackathon showcased incredible ingenuity and problem-solving skills. After a rigorous three-round judging process, including community voting, the final three teams presented their compelling projects.
- Third Place: Volt Control 🥇
- Second Place: Runway Ops 🥈
- First Place: PlanPass AI 🥇🥇🥇
This event truly underscored the power of agentic AI and robust data platforms like MongoDB in driving innovation and creating tangible solutions for complex global challenges. Congratulations to all the participants for their hard work and brilliant ideas! 🎉