Turn any electronic component datasheet into a risk-first engineering summary — adaptive sections, bring-up watch-outs, and decision-ready pros / cons / red flags.
Author: Arshia Keshvari (@TeslaNeuro)
·
License: MIT
Drop in a PDF datasheet. The app extracts text locally in your browser, then asks an LLM to produce a structured summary written like an experienced electronics engineer — focused on risk, bring-up, and design watch-outs.
Output adapts to the part: IC, sensor, power device, connector, module, equipment, and more. Irrelevant sections are omitted; component-specific extras can appear when they matter.
| 🆓 Zero-setup default | Runs on Puter.js — no API key, no server. First analysis signs into a free Puter account; users pay only for their own usage. |
| 🦙 Local / private mode | One-click Ollama preset. Models stay on your machine. |
| 🔑 Bring your own key | OpenAI, OpenRouter, or any OpenAI-compatible endpoint (Groq, Together, vLLM, LM Studio, …). Key stays in localStorage. |
| 📕 Local PDF parsing | pdfjs-dist in the browser. The file never leaves your machine — only extracted text is sent to the model. |
| 🧩 Adaptive sections | Pinout, curves, etc. appear only when relevant; custom extra sections for sensors, SMPS, connectors, and more. |
| Risks & design watch-outs and Decision summary are always present, with explicit red flags. | |
| 📤 Exports | Copy or download as Markdown or JSON. |
| 🪶 No backend | Pure static SPA — host anywhere. |
Nine adaptive sections (omit what doesn’t apply; add extras when useful):
| # | Section | When |
|---|---|---|
| 1 | Component Identification | Always |
| 2 | Absolute Maximum Ratings | If applicable |
| 3 | Recommended Operating Conditions | If applicable |
| 4 | Electrical / Performance Characteristics | If applicable |
| 5 | Pinout / Interface / Connections | If applicable |
| 6 | Recommended Circuits / Application Notes | If applicable |
| 7 | Risks, Caveats & Design Watch-outs | Always |
| 8 | Alternatives & Cross-References | If known |
| 9 | Summary for Decision-Makers | Always (pros · cons · red flags) |
Examples of model-added extras: Calibration (sensor), Safety / Isolation (SMPS), Mating / IP rating (connector).
npm install
npm run devOpen http://localhost:5173 → drop a PDF → Analyse.
The first Puter call opens a sign-in popup; later runs are silent.
Four steps from empty screen to a decision-ready summary:
Drag a PDF into the dropzone (or click to browse). Nothing is uploaded to an app server — parsing stays in your browser.
With a file selected, click Analyse Datasheet. Use Settings first if you want Ollama, OpenAI, OpenRouter, or a custom endpoint instead of Puter.
The app extracts text page-by-page, then streams the model response. You can cancel anytime.
Browse adaptive sections (identification, ratings, risks, decision summary, …). Copy or download as Markdown or JSON.
Open Settings to switch providers:
| Provider | Key? | Suggested models | Notes |
|---|---|---|---|
| Puter | No | claude-sonnet-4-5, gpt-5.4-nano, gpt-5.2-chat, gemini-2.5-flash |
Keyless; user-pays via Puter |
| Ollama | No | qwen3.6:27b, gemma4:12b, qwen3-coder:30b, llama3.3 |
100% local — see CORS below |
| OpenAI | Yes | gpt-5.6-luna, gpt-5.6-terra, gpt-5.4-nano |
Bring your own key |
| OpenRouter | Yes | anthropic/claude-sonnet-4.5, google/gemini-2.5-flash, openai/gpt-5.4-nano |
One key, many models |
| Custom | Yes | Any OpenAI-compatible model | Groq, Together, vLLM, LM Studio, … |
Prefer models that support JSON mode (response_format: json_object). Puter routes may not always honour it, so the prompt also demands a single JSON object and the parser recovers JSON from prose when needed.
1. Install & pull a model
ollama pull qwen3.6:27bModels in the ~12B+ range (or strong MoE equivalents) work best for structured extraction. Smaller 3–8B models often skip required JSON fields.
| Tier | Models | Rough fit |
|---|---|---|
| Recommended | qwen3.6:27b, gemma4:12b |
Best balance for datasheet JSON |
| Coding / long context | qwen3-coder:30b, gpt-oss:20b |
Strong instruction following on 16–24 GB cards |
| Larger rigs | llama3.3:70b, gemma4:31b |
Higher quality when VRAM allows |
| Light / fallback | qwen3:8b, phi4-mini |
Faster, but more incomplete JSON |
2. Enable CORS so the browser can call Ollama (OLLAMA_ORIGINS before starting Ollama):
| OS | Setup |
|---|---|
| macOS | launchctl setenv OLLAMA_ORIGINS "*" → quit & reopen the Ollama menu-bar app |
| Linux | sudo systemctl edit ollama.service → add Environment="OLLAMA_ORIGINS=*" → systemctl restart ollama |
| Windows | System env var OLLAMA_ORIGINS=* → restart Ollama |
Tighter option: OLLAMA_ORIGINS=http://localhost:5173,http://localhost:4173.
3. In the app → Settings → Ollama (local) → confirm base URL (http://localhost:11434/v1) → pick model → Save.
npm run build
npm run previewdist/ is a fully static SPA — host on GitHub Pages, Netlify, Cloudflare Pages, S3, or any static host.
- PDFs are parsed in your browser.
- There is no app backend. Extracted text goes straight from the browser to your chosen provider:
- Puter → Puter’s API → underlying model vendor
- BYOK / Ollama → your configured endpoint; API keys live only in
localStorage
| Topic | Detail |
|---|---|
| Large PDFs | ~180 000 characters max; UI flags truncation and the model notes assumptions |
| Scanned PDFs | Image-only files yield no text — OCR first or use a text-based datasheet |
| Safety-critical use | Always cross-check against the original datasheet |
Vite · React · TypeScript · Tailwind CSS
pdfjs-dist · react-markdown · remark-gfm
Puter.js v2 · OpenAI-compatible chat completions
docs/screenshots/ # Workflow guide images for this README
src/
├── App.tsx # Upload → extract → analyse → results
├── components/ # Dropzone, Settings, Results, UI chrome
└── lib/
├── pdf.ts # Local PDF text extraction + hashing
├── llm.ts # Puter / OpenAI-compatible client
├── prompt.ts # Risk-first system & user prompts
├── cache.ts # IndexedDB extraction & analysis cache
└── … # Types, storage, export, error logging
Arshia Keshvari (@TeslaNeuro)
Created and maintained as an open-source tool for electronics engineers who want faster, safer datasheet triage.
Released under the MIT License.
Copyright © 2026 Arshia Keshvari (@TeslaNeuro)