CLI runtime
One language for instruments
Machine commands, protocols, data extraction, reports, and project memory share the same inspectable PUDA boundary.
Physical Unified Device Architecture
Give AI a hand in the physical world
PUDA connects AI agents to instruments, robots, datastreams, approvals, and audit trails — turning physical experiments into programmable, trustworthy workflows.
AI4Sci beachhead / AI-native hardware / trustworthy physical datastreams
The next wave of AI is not just perception. It is interaction with the physical systems that produce scientific truth.
The PUDA system
The connective tissue between agents and machines.
PUDA is an execution and audit boundary for physical work: AI agents can reason, Genies can coordinate, but every state-changing action remains validated, logged, and reportable.
Agent layer
Genies that plan and coordinate
Coord, First, Bio, and future instrument Genies route work, explain intent, and hand off tasks without bypassing validation.
Trust layer
Approvals, locks, and audit trails
Physical actions pass through capability gates, operator policy, execution logs, post-checks, and durable report artifacts.
Demo slot
Reserved for the PUDA video.
A cinematic space is already in the layout. When the lab video is uploaded, the placeholder becomes a native player.
video space reserved
PUDA lab demo video
A short lab walkthrough will show PUDA routing agent intent through validation, instrument execution, measurement data, and report artifacts.
demo coming soon
Infrastructure loop
From intent to experiment to evidence.
AI agents
PUDA validation
Instrument skills
Physical datastream
Report ledger
Genies
A lab team of Genies, each with a job.
Genies give operators a voice-native, agentic interface to the lab. They can plan, speak, coordinate, and learn patterns — but hardware remains behind PUDA policy.
Coord
routing, planning, approvals
Maintains autonomy envelopes, invariants, shared context, and final verification before results go out.
First
deck, liquid, electrode prep
Owns bounded preparation steps such as deck checks, transfers, and electrode positioning through PUDA.
Bio
electrochemistry measurement
Runs CV-style measurement workflows, interprets signals, and returns markdown reports plus plots.
Gigi / GC
future instrument Genie
A planning slot for chromatography and additional machines as the tool network expands.
Use cases
Physical tasks become programmable environments.
Run AI4Sci experiments
Give AI systems a physical endpoint: propose, act inside a bounded envelope, measure feedback, and update the next move.
Automate flexible SOPs
Turn evolving lab operations into executable protocols that can be previewed, checked, approved, and repeated.
Evaluate AI scientists
Make physical tasks benchmarkable with transparent attempts, rewards, failures, measurement data, and scientific memory.
12-month roadmap
3 months
Deploy the wedge
Integrate the first 10 tools and land the beachhead deployment.
6 months
Expand the wedge
Agentic SOPs, reusable skills, and early ecosystem expansion.
9 months
Enterprise validation
Security and IP middleware plus enterprise proof-of-concept workflows.
12 months
Scalable deployment
Scale toward 500 tools and the AI4Sci integration layer.
Team
Built for AI4Sci.
A technical team and scientific advisors building the bridge between AI systems and real-world experimentation.
ZR
Zekun Ren
AI Expert
KH
Kedar Hippalgaonkar
Prof. NTU AI4Sci
KN
Konstantin Novoselov
Scientific Advisor
Partner with us
Build the physical AI infrastructure layer.
Whether you are integrating lab instruments, evaluating AI scientists, or converting SOPs into reliable agentic workflows, PUDA is designed to make action trustworthy.
Contact PUDA