01
Product & UX
Designed the core loop, Weave chat, research memory timeline, and extension onboarding across VS Code, Cursor, and Windsurf.
- Product
- Brand
- Full case study
- 2026
Noteweave
AI systems for reliable technical R&D — an IDE extension that automates the idea-to-experimentation loop with verifiable research memory.
Long story short
I co-founded Noteweave and lead product design for an AI research assistant built as an IDE extension — 100+ users in six weeks
On a three-founder team at Founders Inc., I own UI/UX for Weave — the agent layer that guides scientists from context to hypothesis, experiment, and export with an immutable research memory.
02
UI systems
Led a full UI revamp — light mode, Blend mode, waiting states, permissions, and webapp token flows.
03
Design engineering
Shipped dozens of UX fixes from a structured extension audit — formatting, numbering, chat scroll, and install flows.
How might we automate the idea-to-experimentation loop for technical R&D — with skepticism built in and a verifiable record of everything?
R&D tooling sits in a gap
Out-of-the-box AI tools for R&D — like Claude Code or Benchling — are either too niche for cross-domain research or too broad to do anything rigorously. Manual R&D cannot keep pace as AI pushes more research than teams can absorb.
Broad-spectrum tools
Frontier models lean expensive without break-even value, or cheap and jack-of-all-trades — both unviable for rigorous R&D.
Niche tools
Domain-specific tools excel in one area but hit a ceiling — constrained to a single field.
Manual research
Teams lose time on irreproducible work while pressure to ship output keeps accelerating.
Community cost
Rigour erodes; significant work stays invisible, unrecognised, and unrewarded.
The insight
R&D in AI/ML, bioinformatics, physics, and pharma looks different on the surface — but the underlying scientific method is the same.
- Understand the problem space
- Survey research in that space
- Check validity of existing work
- Find proven methods and build on top — with valid skepticism
Noteweave is my research partner that never forgets anything. Guides like an advisor, works like an assistant.
Our approach
Build agent systems that improve as frontier models scale in reasoning — with inbuilt skepticism that eliminates unreliable research upfront. Automate the idea-to-experimentation loop with a human in the loop and verifiable records.
Agent architectures that scale with model reasoning
Skepticism as a feature — unreliable research filtered upfront
Human-in-the-loop idea → experiment automation
Immutable IP log and audit trail for every project
Core user loop
Every session follows the same scientific rhythm — from understanding context to exporting findings as research artifacts.
| Stage | Trigger | Output | Weave's role |
|---|---|---|---|
| Understand space | User input | Weave chat output | Intervenes when data ingestion is incomplete |
| Create hypothesis | User approval | Structured hypothesis | Surfaces candidates; user edits or rejects |
| Perform experiments | User approval | Experiment data | Live progress, early failure detection, full logging |
| Analyse results | Run completes | Findings | Summarises results and proposes next steps |
| Export findings | User intent | Research paper + IP log | Blocks export on null data with clear errors |
A failed experiment in Noteweave is never lost work. Every failure enriches the research memory and moves the project forward.
Research memory
Research memory is Noteweave's primary USP — a scientific repository that functions as both knowledge layer and IP logging system. Events are captured automatically, ordered chronologically, and searchable.
Captured automatically
No manual entry — the system logs as the project progresses.
Timeline structure
Progressive event logs with tags — project start, context, hypotheses, experiments, evaluations.
Verified entries
Each log: timestamp, contributor, event description, and testimonial references.
Defensible IP trail
Exportable, immutable audit trail — litigation-ready institutional record.
Weave — research twin
Weave is the AI capability layer between scientific knowledge and the user. Assistive, warm, and motivating — failures are framed as learning moments.
Passive
Observes actions quietly — logs context without interrupting.
Assistive
Surfaces when attention is needed — explains errors, suggests recovery.
Active
Drives progress — hypothesis creation, live experiment updates, artifact generation.
Weave acts autonomously
Suggestions & analysis
Converses in natural language and analyses results.
Research memory
Logs all events without user action.
Draft hypotheses
Generates structured, testable claims from context.
Requires approval
- Running experiments
- Publishing research papers
- Exporting IP logs
- Adding collaborators
Hypothesis & experiments
Hypotheses are structured, testable claims generated by Weave from ingested context — approved by the user before any experiment runs. Every run is logged with environment snapshots, reproducibility scores, and Weave notes.
- Confidence score (0–100) based on supporting evidence in the knowledge base
- Experiment lifecycle: Design → Review → Approved → Running → Analysed
- Failure taxonomy with recovery paths — code, environment, ingestion, and approval errors
Where Noteweave fits
Noteweave complements coding agents like Claude Code — using academic research to propose product improvements today, with solitary coding agents on the roadmap.
| Competitor | Cross-domain R&D | Verifiable IP log | Hypothesis → experiment loop | IDE-native |
|---|---|---|---|---|
| Noteweave | ||||
| Claude Code | ||||
| Benchling | ||||
| ChatGPT / Claude | ||||
| Perplexity |
Progress & traction
100+ users
Extension users
Launched ~6 weeks ago · Founders Inc. residency
- Review system benchmark
- +10% vs Opus 4.6 & GPT 5.4
- Design partners
- 5
- Founding team experience
- 12 yrs AI/ML
- North star (MVP target)
- >3 experiments / project / mo
- Review system outperformed Claude Opus 4.6 and GPT 5.4 at finding technical issues by 10% (technical report)
- Working with 5 design partners in applied AI/ML for feedback loops
- Noteweave takes business problem context and uses academic research to propose product improvements
- Three ex-founders with specialisation in research, engineering, and UI/UX
First 10 minutes
The first session must deliver one moment of value — no onboarding wizard, just a single prompt: What are you working on?
| Minute | What happens | Success signal |
|---|---|---|
| 0–1 | Clean project creation — one prompt | Project created |
| 1–3 | User pastes a paper or describes a problem; Weave confirms understanding | Context accepted |
| 3–6 | Weave surfaces a structured hypothesis | Hypothesis approved |
| 6–10 | First experiment runs or hypothesis is refined in detail | Experiment initiated |
UI/UX — extension & webapp
I led a structured UI/UX audit and revamp across the VS Code extension, webapp, and marketing site — fixing chat behaviour, onboarding, permissions, and visual systems.
Extension chat UX
Smooth auto-scroll, clickable links, fixed numbering (1, 2, 3…), and line breaks after e.g.
Blend mode
Renamed Auto → Blend Mode as default; mode-switching no longer drops answers mid-chat
Waiting states
Added personality — Output Maxing…, Aura Farming…, Rizzing… — with fade instead of abrupt swaps
Onboarding & install
Download CTA moved above plans; workspace-first gating — no chat until a folder is open
- Full light-mode UI revamp across extension and webapp
- Token creation split into separate boxes; auto-delete stale tokens
- Permissions overlay condensed to one-line asks with waiting state
- Paper analysis panel — close button, direct-open on progress bar, 95% max display
- Footer & CTA routes to apps.noteweave.io auth; Slack → Discord
- Welcome greetings with first name; increased question font size by 4px
- File-type icons, hover overlays, and Pro / Pro+ tier labelling
Defensibility
R&D as a moat
State-of-the-art ML models powering Weave's reasoning — continuously improved as frontier models scale.
Litigation as a moat
Automated, immutable IP log trail — defensible legal and institutional record of discovery.
Memory lock-in
Research memory accumulates value over time — leaving means losing continuity.
Learnings & impact
Skepticism is the product
Users trust Weave when it flags unreliable research upfront — not when it generates more.
Design for the full loop
Optimising chat alone misses the value — the timeline, IP log, and export artifacts create retention.
Ship UX fixes in public
A structured audit with Discord feedback turned dozens of small bugs into a cohesive UI system.
Paridhi was my Co-founder in 2 startups, which we scaled to users from 30+ countries. She has phenomenal grit and thinking. You are at a loss if you do not have her in your team!
Ready for next?