HackFW: MADE Challenge

Convergent Technology Hackathon — Building at the intersection of data engineering, computer vision, blockchain, robotics, and agentic AI for manufacturing.

Build a working solution to a real problem: inspect parts, monitor equipment, trace materials, coordinate robots, or help operators make better decisions.

This is a menu of recommended technologies, not a checklist. Choose the smallest stack that supports your demo. You do not need a tool from every category, and you can use technologies beyond this list.

Hardware bonus: Extra points are available for a working solution on a microcontroller, such as an STM32 or Arduino board.

These are suggested combinations, not prebuilt integrations. Choose one option wherever alternatives are listed.

1. Visual quality inspection
Stack: Python + OpenCV + Ultralytics YOLO or Anomalib + FastAPI + PostgreSQL + React
Demo: Detect a defect, save inspection evidence, and route uncertain cases to a person.

2. Manufacturing analytics
Stack: Python + Polars + DuckDB + Parquet + Streamlit; add MotherDuck for shared cloud analytics
Demo: Explore production data and explain yield, downtime, or scrap trends.

3. Knowledge graph / GraphRAG
Stack: Python or TypeScript + Neo4j or HelixDB + an LLM API + React
Demo: Connect machines, parts, manuals, and maintenance records; show the evidence behind answers.

4. Equipment telemetry
Stack: STM32 or Arduino + MQTT + Mosquitto + a Python ingestion service + TimescaleDB or QuestDB + Grafana
Demo: Read a real sensor, chart its history, and trigger an alert.

5. TinyML on a microcontroller
Stack: A supported STM32 or Arduino board + Edge Impulse or LiteRT for Microcontrollers + a sensor
Demo: Run a small model on the board and report measured latency and memory use.

6. Operations copilot
Stack: Next.js + AI SDK + PostgreSQL/pgvector + an LLM API; deploy on Vercel
Demo: Search approved procedures, cite sources, and draft an action for human approval.

7. Supply-chain traceability
Stack: TypeScript + viem/wagmi + a Base testnet contract + PostgreSQL + object storage
Demo: Link an off-chain inspection record to an on-chain hash and verify provenance.

8. Robotics / digital twin
Stack: ROS 2 + Gazebo + MoveIt 2 + Python/C++; add micro-ROS for supported microcontrollers
Demo: Simulate a robot task, inspect its state, and demonstrate a controlled hardware interaction.

Deployment starting point: For a conventional API or container, consider Railway, Render, Cloud Run, Azure Container Apps, or ECS/Fargate. For a web frontend, consider Vercel, Netlify, or Cloudflare. Use a local machine, Jetson, or GPU service when your vision/model workload requires it. These platforms have different runtime and persistence limits; check your workload before choosing.

• TypeScript — Typed web apps, APIs, and AI integrations. Docs
• Node.js — JavaScript services and the npm ecosystem. Docs
• Bun — JavaScript/TypeScript runtime, package management, and builds. Docs
• Deno — JavaScript/TypeScript runtime with permission controls. Docs
• Python — Computer vision, data analysis, ML, and automation. Docs
• uv — Python environments, dependencies, and reproducible project setup. Docs
• Go — Backend services, telemetry ingestion, and networking. Docs
• Rust — Systems software, device services, and performance-sensitive backends. Docs
• C / C++ — Microcontroller firmware, robotics, and native inference. GNU C · C++
• Zig — Systems programming and native tooling. Docs

• React — Interactive dashboards and operator interfaces. Docs
• Next.js — Full-stack React applications. Docs
• Vite — Frontend development server and build tooling. Docs
• TanStack Start / Router — Full-stack React applications and typed routing. Start · Router
• TanStack Query — Fetching, caching, and synchronizing server data. Docs
• TanStack Table / Form — Data tables and form state. Table · Form
• Tailwind CSS — Utility-based styling. Docs
• shadcn/ui — Customizable interface components. Docs
• FastAPI — Python APIs for data and ML services. Docs
• Hono — Lightweight TypeScript APIs across supported runtimes. Docs
• Streamlit — Data apps and analytics demos in Python. Docs
• Gradio — Model demos with image, audio, and text inputs. Docs

Use these for application records: users, assets, work orders, inventory, and inspection results.

• PostgreSQL — General-purpose transactional data and SQL. Docs
• SQLite — Embedded applications and simple local persistence. Docs
• MySQL — Transactional applications and existing MySQL integrations. Docs
• Turso — SQLite-compatible databases, hosted deployments, and sync options. Docs

These services package hosting or backend capabilities around a database; they are not all separate database engines.

• Supabase — PostgreSQL with authentication, storage, and realtime APIs. Docs
• Neon — Managed PostgreSQL with database branching. Docs
• PlanetScale — Managed PostgreSQL or Vitess/MySQL. Docs
• Convex — Reactive application data, server functions, and subscriptions. Docs
• Cloudflare D1 — Managed SQL with SQLite semantics for Workers applications. Docs

Use these for historical analysis, event aggregation, files, and larger datasets. DataFrame libraries and file/table formats support databases; they are not interchangeable with them.

• DuckDB — Embedded analytical SQL engine for CSV, Parquet, and other data sources. Docs
• MotherDuck — Managed analytics platform extending DuckDB into the cloud. Docs
• ClickHouse — Column-oriented database for event analytics and aggregations. Docs
• Polars — DataFrame library for transforming and analyzing data. Docs
• Apache Arrow — Columnar in-memory format and data interchange tooling. Docs
• Apache Parquet — Columnar file format for analytical datasets. Docs
• DuckLake — Lakehouse table format using a SQL metadata catalog and Parquet data. Docs
• Apache Iceberg — Table format for analytical data in object storage. Docs
• Delta Lake — Transactional table format for data lakes. Docs
• BigQuery — Managed analytical warehouse on Google Cloud. Docs
• Snowflake — Managed data warehousing and analytics. Docs
• Databricks — Managed data engineering, lakehouse analytics, and ML. Docs

Practical default: Start with DuckDB + Parquet for a file-based analytics demo. Add a shared service or catalog when multiple users or processes need coordinated access. Keep a transactional database for application workflows when needed.

Use these when relationships are central: supplier networks, bills of materials, machine dependencies, process lineage, and connected maintenance knowledge.

• Neo4j — Property graphs, Cypher queries, graph algorithms, and GraphRAG. Docs
• HelixDB (helix-db) — Graph traversal and vector search in one engine for connected AI retrieval. Docs
• Memgraph — Graph applications, Cypher queries, and graph analytics. Docs
• FalkorDB — Graph queries and knowledge-graph retrieval for AI applications. Docs
• ArangoDB / Arango — Graph and document data queried together. Docs
• SurrealDB — Multi-model applications combining records, relationships, and vector search. Docs

Choosing a graph database: Neo4j is a useful starting point for teams wanting Cypher tutorials and graph tooling. HelixDB is worth exploring when combining relationships with semantic retrieval is central to the project. The products use different APIs and query models; confirm SDK support and deployment requirements early.

Use these for similarity search across manuals, text, images, and embeddings. Add an embedding model and an ingestion process; a vector index does not create those for you.

• pgvector — Vector similarity search inside PostgreSQL. Docs
• Qdrant — Vector retrieval with metadata filtering. Docs
• Chroma — Embedding storage and retrieval for AI applications. Docs
• Weaviate — Vector and keyword search with metadata filtering. Docs
• Pinecone — Managed vector search. Docs
• Milvus — Vector search for larger collections. Docs
• LanceDB — Embedded vector/hybrid retrieval and multimodal data workflows. Docs

Practical default: If you already use PostgreSQL, evaluate pgvector before adding another database. A graph database with vector search may also cover both needs.

• TimescaleDB — PostgreSQL-based time-series data and SQL analysis. Docs
• QuestDB — Timestamped sensor/event data and time-series SQL. Docs
• InfluxDB — Telemetry, measurements, and time-series workloads. Docs
• MongoDB — Document-oriented application records. Docs
• Amazon DynamoDB — Managed key-value and document workloads on AWS. Docs
• Redis — Caching, counters, streams, and pub/sub. Docs
• Valkey — In-memory key-value data, caching, and messaging. Docs

Store images, video, model files, exports, and raw datasets here. Keep searchable metadata and access permissions in your application database.

• Cloudflare R2 — Object storage with an S3-compatible API. Docs
• Amazon S3 — Object storage on AWS. Docs
• Google Cloud Storage — Object storage on Google Cloud. Docs
• Azure Blob Storage — Object storage on Azure. Docs
• OCI Object Storage — Object storage on Oracle Cloud. Docs

• OpenCV — Camera capture, image processing, calibration, and classical vision. Docs
• Ultralytics YOLO — Object detection, segmentation, tracking, and counting. Docs
• Anomalib — Visual anomaly detection, including unusual defects or surface changes. Docs
• Segment Anything (SAM) — Promptable segmentation; select a model suited to your hardware. Project
• PaddleOCR — Reading labels, serial numbers, forms, and document text. Docs
• PyTorch / torchvision — Training, adapting, and evaluating vision models. PyTorch · torchvision
• Hugging Face Transformers — Vision-language models and pretrained model integrations. Docs
• Open3D — Point clouds, depth data, and 3D geometry. Docs

• Roboflow — Dataset preparation, model workflows, and vision deployment. Docs
• CVAT — Annotating images and video with boxes, masks, and other labels. Docs
• FiftyOne — Exploring datasets and inspecting model errors. Docs

• ONNX Runtime — Running exported models across supported CPU/GPU platforms. Docs
• NVIDIA TensorRT — Optimizing inference on supported NVIDIA GPUs. Docs
• OpenVINO — Optimizing inference across supported Intel hardware. Docs
• Edge Impulse — Collecting sensor data, training models, and deploying to supported devices. Docs
• LiteRT for Microcontrollers — Small-model inference on memory-constrained microcontrollers. Docs
• STM32Cube.AI / X-CUBE-AI — Converting and deploying supported neural networks on STM32. Product & resources

Demo guidance: Start with a camera or recorded sample, a pretrained baseline, and a small labeled evaluation set. Show latency, false positives, and missed defects. Keep train/test examples separate, including nearby video frames. Use approved footage and review software, dataset, and model-weight licenses before sharing or deploying; Ultralytics provides specific licensing guidance.

Hardware distinction: Raspberry Pi Linux computers and NVIDIA Jetson systems can host larger vision pipelines. Microcontrollers such as STM32 and many Arduino boards have much tighter memory and compute limits. Do not assume a desktop detector or vision-language model can run on the board; verify the exact board, model, and runtime combination.

• STM32 / STM32Cube — Microcontroller firmware, sensors, and peripheral integration. Docs
• Arduino — Accessible microcontroller prototypes and sensor integrations. Docs
• ESP32 / ESP-IDF — Connected microcontroller applications; features vary by chip. Docs
• PlatformIO — Embedded build, dependency, and upload workflows. Docs
• MicroPython — Python on supported microcontrollers. Docs
• Zephyr — Real-time operating system for embedded devices. Docs
• Raspberry Pi — Linux edge gateways and camera applications; Pico is a separate MCU family. Docs
• NVIDIA Jetson — GPU-accelerated edge vision and robotics. Docs

• ROS 2 — Robot communication, components, and tooling. Docs
• micro-ROS — Connecting supported microcontrollers to a ROS 2 system. Docs
• Gazebo — Robot and environment simulation. Docs
• MoveIt 2 — Motion planning and manipulation. Docs
• NVIDIA Isaac Sim — Robotics simulation and synthetic data generation. Docs
• Foxglove — Inspecting robot messages, recordings, and visualizations. Docs

Use simulation or an isolated demonstration setup for control experiments. Keep hardware interlocks and operator approval outside the AI model's control path.

• MQTT / Eclipse Mosquitto — Lightweight device messaging through an MQTT broker. Docs
• EMQX — MQTT broker and device-data integration. Docs
• NATS — Service messaging, pub/sub, and request/reply. Docs
• NATS JetStream — Persistent streams, consumers, replay, and acknowledgments. Docs
• NATS KV — Key-value buckets backed by JetStream. Docs
• Apache Kafka — Distributed event streaming. Docs
• Redpanda — Kafka-compatible event streaming platform. Docs
• WebSockets / Server-Sent Events — Live browser updates; SSE is server-to-client. WebSockets · SSE
• Node-RED — Visual flows connecting devices, APIs, and events. Docs
• OPC UA / open62541 — Industrial data interoperability and equipment integration. Docs
• Apache Spark — Distributed data processing when the workload warrants it. Docs
• dbt — SQL transformations and data-model checks. Docs
• Dagster — Orchestrating data assets and pipelines. Docs

Delivery semantics: Plan for retries and duplicate messages. Use event IDs and idempotent writes; broker guarantees do not automatically make external database writes or physical actions happen exactly once.

Choose a provider based on your team's familiarity, runtime, hardware requirements, and available budget. No credits, free tiers, sponsorships, or GPU availability are implied by inclusion.

• Cloudflare — Workers/Pages for web apps; Durable Objects for stateful coordination; Queues for asynchronous work. Workers · Pages · Durable Objects · Queues
• Vercel — Frontend/full-stack deployments, preview URLs, and Functions. Docs
• Netlify — Web deployments, deploy previews, and Functions. Docs
• Railway — Deploying APIs, containers, workers, and database services. Docs
• Render — Web services, background workers, scheduled jobs, and managed PostgreSQL. Docs
• Fly.io — Applications on Machines with regional placement and volumes. Docs

• AWS — Lambda; ECS/Fargate for containers; EC2; RDS; S3; IoT Core. Docs · Fargate
• Google Cloud (GCP) — Cloud Run; Compute Engine; Cloud SQL; BigQuery; Cloud Storage; Pub/Sub. Docs
• Microsoft Azure — Container Apps; Functions; Virtual Machines; Azure Database for PostgreSQL; Blob Storage; IoT Hub. Docs · Container Apps
• Oracle Cloud Infrastructure (OCI) — Compute; Container Instances; OKE; Functions; Object Storage; database services. Docs
• DigitalOcean — Droplets, App Platform, managed databases, and Spaces object storage. Docs
• Hetzner — Cloud VMs, networking, volumes, and self-managed deployments. Docs

• Modal — Python-based CPU/GPU jobs and serverless model endpoints. Docs
• Runpod — GPU Pods and serverless inference deployments. Docs
• Lambda — GPU instances and clusters for training or inference. Docs

AWS, Google Cloud, Azure, and OCI also have GPU offerings. Check quotas, region availability, model memory requirements, and shutdown behavior before the event. Use recorded inputs or a smaller local model as a fallback.

Retained from the original list for teams using the Cloudflare ecosystem.

• R2 / D1 — Object storage and managed SQL; see the database sections above. R2 · D1
• Workers KV — Distributed key-value reads; account for consistency behavior. Docs
• Vectorize — Managed vector indexes. Docs
• Workers AI — Hosted model inference. Docs
• AI Gateway — AI request routing, monitoring, and supported caching controls. Docs
• DNS / WAF — DNS management and web traffic filtering. DNS · WAF

• Docker / Docker Compose — Packaging services and running a local multi-service demo. Docker · Compose
• Kubernetes / K3s — Container orchestration; K3s provides a lighter distribution. Kubernetes · K3s
• Terraform / OpenTofu — Declarative infrastructure provisioning. Terraform · OpenTofu
• Pulumi — Infrastructure definitions in programming languages. Docs
• GitHub Actions — Automated builds, tests, and deployments. Docs
• OpenTelemetry — Instrumenting traces, metrics, and logs. Docs
• Prometheus / Grafana — Metrics collection and operational dashboards. Prometheus · Grafana
• Sentry — Application errors and performance diagnostics. Docs
• Tailscale — Private connectivity between development and edge machines. Docs

For most hackathon projects, a managed deployment or Docker Compose is sufficient. Add Kubernetes when orchestration itself is relevant or your team already knows it. Store secrets outside source code, protect device endpoints, and set billing alerts.

Vision-language models can interpret images and text, but their descriptions are not a substitute for measured detection accuracy or hardware safety controls.

• Featherless.ai — Hosted open-model inference. Docs
• Together AI — Hosted model inference and supported training workflows. Docs
• Hugging Face — Model discovery, model cards, and hosted/local integrations. Docs
• Ollama — Running supported models locally. Docs
• vLLM — Serving supported models with GPU inference infrastructure. Docs
• Cloudflare Workers AI — Hosted inference through Cloudflare. Docs

Pick one framework if your project needs orchestration. A direct model API call with a small number of tools may be enough.

• Mastra — TypeScript agents and workflows. Docs
• Agno — Agent applications and multi-agent workflows. Docs
• LangChain — Model/tool integrations and agent building blocks. Docs
• LangGraph — Stateful agent workflows with persistence and control flow. Docs
• CrewAI — Role-based agent workflows. Docs
• Microsoft Agent Framework — Python/.NET agents and orchestration; successor path for AutoGen users. Docs
• LlamaIndex — Ingestion and retrieval workflows over application data. Docs
• Pydantic AI — Typed Python agents and structured outputs. Docs

• CopilotKit — AI interfaces embedded in applications. Docs
• AI SDK — TypeScript model calls, streaming interfaces, and tool integrations. Docs
• Instructor — Structured model outputs validated against schemas. Docs
• DSPy — Programming and optimizing model-driven pipelines. Docs
• Model Context Protocol (MCP) — Connecting AI applications to tools and data through a shared protocol. Docs
• Langfuse — Tracing and evaluating LLM application behavior. Docs
• MLflow — Tracking ML experiments, artifacts, and evaluations. Docs

Give agents only the tools and data they need. Require human approval before physical control actions or changes to important records, and show sources for factual answers.

Use blockchain where shared verification, provenance, or coordination between parties adds value. Keep large files and sensitive manufacturing data off-chain. Start with local development networks or testnets.

• Solidity — EVM smart contracts. Docs
• Ethereum — Smart contracts and settlement. Docs
• Base — Ethereum L2 applications. Docs
• Arbitrum — Ethereum L2 applications and tooling. Docs
• Optimism / OP Stack — Ethereum L2 applications and OP Stack development. Docs
• Polygon PoS — EVM applications on the Polygon PoS network. Docs

• Solana — Programs and applications in the Solana ecosystem. Docs
• Sui — Move-based applications and object-centric data. Docs
• Aptos — Move-based smart contracts. Docs
• NEAR — Smart contracts and applications on NEAR. Docs
• Avalanche — EVM applications and custom Avalanche L1s. Docs
• Cosmos SDK / IBC — Application-specific chains and interoperability. SDK · IBC
• Bitcoin — Bitcoin transactions and protocol integrations. Docs
• Stacks — Smart contracts and applications connected to Bitcoin. Docs
• Lightning — Bitcoin payment-channel applications. Docs

• Foundry — Solidity development, local networks, and testing. Docs
• Hardhat — Ethereum development and testing. Docs
• Anchor — Solana program development. Docs
• thirdweb — Web3 SDKs and application infrastructure. Docs
• wagmi — React hooks for Ethereum applications. Docs
• viem — TypeScript Ethereum interfaces. Docs
• ethers.js — JavaScript/TypeScript Ethereum library. Docs
• OpenZeppelin Contracts — Reusable smart-contract components. Docs

Solve real problems and test whether your solution meets a user's needs. Focus on a clear workflow, a working demonstration, and the skills your team wants to build. Visit the HackFW Discord for ideas from the innovation bounty board, team formation, and support.

HackFW Discord: Join the support community

Documentation links are provided for getting started. Product support, licenses, runtime compatibility, quotas, and availability can change; confirm the details for your selected stack.