
AI & knowledge systems
Evidence, retrieval, agents, and the paths that connect a question to a grounded answer.
Wiki
Concepts from the projects and writing, rebuilt as generated visual worlds with live HTML simulations, source-grounded explanations, and worked examples.
Explore the systems

Evidence, retrieval, agents, and the paths that connect a question to a grounded answer.

Motion, sensing, control loops, and plans made visible as a field of forces and trajectories.

Dense biological and visual signals resolved into structure, geometry, and interpretable layers.

Combinatorial structure, time series, transit flows, and compute systems sharing one visual grammar.
40 of 40 explainers
A* finds paths by balancing the cost already paid with a heuristic estimate of what remains.
Separate hardware access from control logic so one robot program can run with real, simulated, or replayed inputs.
Use known field landmarks to produce time-stamped, uncertainty-aware pose measurements that complement wheel odometry and a gyro.
Reliable API clients treat failure as a flow-control signal, not just an exception to retry immediately.
Transit demand becomes legible only after irregular vehicle observations are validated, deduplicated, aligned, and aggregated with their coverage visible.
Course scheduling is a constraint problem disguised as a registration chore.
A course knowledge base turns files, assignments, and notes into searchable objects with source context.
A dual-arm robot needs more than an action list; it needs a schedule that respects resources, precedence, and collision risk.
TAMP connects symbolic task choices to geometric motion checks so a robot plan is both logical and physically executable.
Elementary symmetric functions collect every square-free product of a fixed degree into one permutation-invariant polynomial.
A think envelope turns an agent answer into a reviewable object with citations, gaps, conflicts, and next actions.
The same joystick vector means different motion depending on the coordinate frame in which it is interpreted.
Economic plotting is a data-contract problem before it is a charting problem: identity, frequency, units, adjustment, missingness, and revision status determine the claim.
A racing game feels right when engine force, steering, friction, suspension, and track design reinforce each other.
A heuristic is a guess about remaining cost; the right guess depends on how movement is allowed.
A Hi-C map turns an ensemble of ligation events into a binned view of how often genomic regions were near one another.
A multi-agent backend is a graph of roles, tools, state, and handoffs, not just a pile of prompts.
A feasibility predictor estimates which symbolic actions are likely to survive expensive motion checks.
When robot physics is messy, a measured lookup table can beat a beautiful but incomplete equation.
Markdown becomes a lightweight knowledge graph when notes carry typed metadata and relationships with enforceable meaning.
MCP standardizes how AI hosts connect to servers that expose tools, resources, and prompts; the host still owns workflow policy and safety.
Small-model engineering chooses a Pareto point across task quality, latency, throughput, memory, and operational constraints.
A boundary call is stronger when its insulation pattern persists across justified window sizes, resolutions, and replicates.
A data agent becomes reviewable when a plain-language request leaves behind executable code, explicit assumptions, output data, and validation evidence.
Autonomous routing combines preplanned paths, event markers, robot constraints, and live pathfinding.
Semantic memory lets an assistant retrieve relevant past context instead of treating every conversation as isolated.
PID control pushes a mechanism toward a setpoint by reacting to present, accumulated, and changing error.
A q-analogue recovers a classical object while using powers of q to record additional structure.
Retrieval-augmented generation turns messy course files into ranked evidence before an AI tutor answers.
A polling loop turns a changing vehicle feed into an auditable historical time series.
A research repository can preserve the plans, source definitions, assumptions, transformations, artifacts, and verification records needed to reconstruct a result.
SGBM estimates stereo disparity by combining local matching evidence with smoothness costs accumulated along multiple image paths.
An HPC training run combines a resource request, executable batch script, recorded environment, and durable outputs.
An AI wiki is useful when every explanation remains tied to the material that produced it.
Calibrated stereo turns horizontal correspondence shifts into metric depth, with uncertainty that grows as disparity shrinks.
A swerve drivetrain independently steers and drives each wheel module so the chassis can translate and rotate at the same time.
Designing an enhancer is a constrained propose–predict–filter–test loop, not a request for the sequence with the largest model score.
The finite q-binomial theorem expands a product whose choices carry position-dependent q-weights.
Keyword search finds exact words; vector search finds nearby meaning. Strong study systems usually need both.
Velocity control is about holding speed under load, not merely sending a motor percentage.