Agentic AI for the real world

See how an agentic AI platform builds a Japanese medical transcription service, validating terms against a lexicon for accuracy.

GraphN Agentic Cursor Agentic AI models fuctions Vector Stores S3 enabled storage

Overview

Lightning AI has created an Agentic AI platform that orchestrates and executes workflows aimed towards solving real world problems. Using this platform I created a medical transcription service that operates in native japanese and cross references with a medical lexicon to validate technical accuracy of medical terms used. This alone is a business model that supports entire companies in the wider medical IT field.

Tech stack

GraphN

GraphN is a node-based visual programming environment by Polygonflow that lets 3D artists build custom, procedural tools for Unreal Engine and Maya without writing code.

Developed by Polygonflow, GraphN (which has evolved into the world-building toolset Dash) solves a major bottleneck for 3D artists and level designers by removing the need for traditional coding. The platform features an intuitive, node-based interface that allows creators to build custom procedural tools, automate repetitive tasks, and scatter assets like foliage or rocks directly in the viewport. By wrapping complex logic into simple, parametric controls, GraphN empowers non-programmers to create non-destructive workflows across Unreal Engine 5 and Autodesk Maya, keeping the focus entirely on artistic decision-making.

https://www.polygonflow.io

View projects
Agentic

Autonomous AI systems that move beyond chat to execute multi-step workflows and achieve complex goals with minimal human oversight.

Agentic technology represents the shift from generative outputs to autonomous action. Unlike standard LLMs that wait for prompts, agentic systems (built on frameworks like CrewAI or LangGraph) orchestrate multiple specialized agents to solve end-to-end problems. A typical stack might involve a Researcher agent scraping 50+ sources via Firecrawl and a Writer agent drafting a report in Google Docs: all coordinated by a central reasoning engine. This paradigm is already driving a 40% increase in operational efficiency for early adopters by automating high-context tasks like insurance claims processing and software debugging. It is not just about talking: it is about doing.

https://www.agentic.ai

View projects
Cursor

The AI-native code editor designed for high-velocity development through deep LLM integration.

Cursor is a fork of VS Code that embeds AI directly into the development workflow while maintaining full extension compatibility. It leverages models like Claude 3.5 Sonnet and GPT-4o to power features such as Cmd+K for inline edits and Cmd+L for codebase-wide chat. By indexing local files, Cursor provides precise context for its predictive 'Tab' completions and multi-file 'Composer' mode. This setup allows engineers to move from high-level intent to functional code without leaving the editor or losing context.

https://cursor.com

View projects
Agentic AI

Autonomous AI systems that plan, execute, and adapt multi-step actions to achieve complex, high-level objectives with minimal human oversight.

This is the next generation of AI: a proactive digital colleague. Agentic systems use large language models (LLMs) as the core reasoning engine, orchestrating specialized AI agents to break down a broad goal (like 'optimize the supply chain') into actionable, real-time steps. Unlike traditional automation, which follows rigid rules, Agentic AI employs a continuous loop of perception, planning, and action, leveraging tools and memory to adapt its strategy on the fly. Gartner predicts that by 2028, one-third of all software will include this technology, automating up to 15% of daily tasks and creating scalable digital labor. This capability fundamentally shifts AI from a content generator to an independent, goal-driven executor.

https://www.ibm.com/topics/agentic-ai

View projects
models

Models: AI's core, transforming data into actionable intelligence.

Models are the computational brains of AI, converting raw data into predictive and generative power. They range from simple linear regressions to complex neural networks like Google's LaMDA and OpenAI's GPT-4, driving applications from fraud detection at financial institutions to personalized recommendations on e-commerce platforms. These algorithms, trained on vast datasets, identify patterns and make informed decisions, continuously learning and adapting to new information. Their effective deployment requires robust data pipelines, significant computational resources (e.g., GPUs), and meticulous validation to ensure accuracy and mitigate bias, ultimately delivering tangible business value across industries.

https://www.ibm.com/topics/artificial-intelligence

View projects
fuctions

Execute event-driven code in a serverless environment: Google Cloud Functions scales automatically, managing infrastructure so you focus on logic.

Google Cloud Functions provides a serverless execution environment for your code. It responds to events (e.g., HTTP requests, Cloud Storage changes, Pub/Sub messages), executing your logic without provisioning or managing servers. This allows developers to build scalable, event-driven applications and microservices efficiently. Pay only for compute time consumed, optimizing costs. Supported languages include Node.js, Python, Go, Java, .NET, Ruby, and PHP.

https://cloud.google.com/functions

View projects
Vector Stores

Vector Stores are specialized databases designed to efficiently store high-dimensional numerical embeddings, enabling rapid semantic similarity search for AI applications like Retrieval-Augmented Generation (RAG).

Vector Stores (or Vector Databases) are the core infrastructure for modern semantic search: they manage multi-dimensional vector embeddings, which are numerical representations of unstructured data (text, images, audio). Unlike traditional databases, they perform similarity searches—not keyword matching—using algorithms like HNSW (Hierarchical Navigable Small World) to find the closest vectors in the embedding space. This capability is critical for powering Large Language Model (LLM) applications, specifically in RAG systems, where external, up-to-date knowledge must be retrieved accurately. Key players in this space include open-source options like Milvus and Chroma, and managed services such as Pinecone and Weaviate.

https://milvus.io/

View projects
S3 enabled storage

S3 enabled storage offers scalable, secure, and cost-effective object storage for any data, anytime, anywhere.

S3 enabled storage provides a robust solution for storing and retrieving any amount of data from anywhere on the web. It's designed for 99.999999999% (11 nines) of durability, making it ideal for critical data like backups, archives, and big data analytics. With features like versioning, lifecycle management, and multiple storage classes (e.g., S3 Standard, S3 Intelligent-Tiering, S3 Glacier), you optimize costs while meeting performance needs. Integration with AWS services (e.g., EC2, Lambda) and third-party applications is seamless, ensuring flexible and efficient data workflows.

https://aws.amazon.com/s3/

View projects

Agentic AI for the real world

Related talks