Electronics lab
Generate and inspect circuits through KiCAD and Icarus Verilog workflows.
Hackathon / AI education
Synapse
An agent-based learning lab for electronics, mathematics, physics, and chemistry.
The learning model
Technical learning works better when explanation and experimentation share a workspace.
Students learning electronics, mathematics, physics, and chemistry often jump between explanations, simulators, and disconnected tools.
Synapse coordinates specialist agents with established technical software so a question can become a circuit, simulation, animation, or molecular view inside one learning flow.
Mathematics lab
Turn abstract explanations into rendered Manim sequences.
Chemistry lab
Explore molecules through RDKit and interactive 3D structures.
Specialist agents
Route each learning task to tools and reasoning suited to its domain.
What I built with the team
A tool runner with an agent in front, not a chat UI pretending to be a lab.
Our team built Synapse because a general chat window cannot run the electronics, mathematics, and science tools students need to inspect real outputs.
I worked across the specialist-agent flow, frontend, backend coordination, and the integrations that turn a prompt into a circuit, simulation, animation, or molecule view.
We connected agents to KiCAD, Icarus Verilog, Manim, RDKit, 3Dmol, Python, ChromaDB, MongoDB, and S3 during the hackathon build.
Route by domain before choosing a tool
A classifier narrows the problem first. The selected specialist can then prepare inputs and interpret output for the relevant technical tool.
Use real executors
We preferred a KiCAD, Verilog, Manim, or RDKit result over an agent describing what such a result might look like.
Return the artifact with the explanation
The learning value comes from seeing and changing the generated output, not only reading a final answer in chat.
How it works
Agents coordinate proven technical tools instead of imitating them.
The application routes a learning request to a domain agent, which prepares inputs for the relevant simulator or rendering tool and returns the result to the shared workspace.
TypeScriptAgentsKiCADRDKitMongoDBAWS
Use AI to connect learners to real tools, not to replace the tools with plausible text.
I worked across the agent workflows, product interface, backend coordination, and integrations that turned established engineering tools into a coherent learning environment.Where the prototype stopsThe repository still shows hackathon speed: separate backend experiments, uneven setup documentation, and several heavyweight tool dependencies.Generated code and scientific artifacts need domain validation. A successful tool process is not automatically a correct lesson.Running user-directed technical tools in the cloud would require a much stronger sandbox and resource-isolation model.