Skip to content

Bib569/quinone-task-ai

BranchesTags

Repository files navigation

QuantumTask — AI Molecular Property Analyzer

Netlify Status Live Demo License: MIT

A task-oriented, step-by-step molecular analysis tool for quinone derivatives and arbitrary SMILES molecules. Powered by the Groq API with selectable free-plan models (Kimi K2, Llama 3.3 70B, Qwen3 32B, GPT-OSS 120B) and optional thinking mode.

🔗 Live Demo: https://quinone-task-ai.netlify.app
🔗 Companion App (chat-based): QuantumChat

✨ Features

Feature Description
4-Step Workflow Input → Properties → HOMO-LUMO Chart → AI Report
Molecular Descriptors MW, XLogP, TPSA, HBD/HBA, rotatable bonds from PubChem
HOMO-LUMO Bar Chart Interactive Recharts chart comparing all 25 quinones + your molecule
DFT Reference Data Instant lookup against B3LYP/6-31G* computed HOMO/LUMO/gap
k-NN ML Prediction Gap estimates for unknown molecules with confidence scoring (±0.3–0.5 eV)
AI Report Publication-quality chemistry report with method recommendations
QC Methods Library Searchable panel: 20+ methods, 25+ functionals, 18+ basis sets
Export Report Download report as Markdown, PDF, or Word (.docx)
Model Selector Dropdown in header to switch between 4 Groq models (Kimi K2, Llama 3.3 70B, Qwen3 32B, GPT-OSS 120B)
Thinking Mode Toggle beside model selector — enabled for Qwen3 32B; reasoning shown in collapsible panel
Auto-retry Server retries up to 3× on 502/503/429 errors automatically

⚠️ Rate Limits (Groq Free Plan)

Groq's free plan has per-minute token limits per model. If you hit a rate limit (HTTP 429):

  1. Wait 10–30 seconds and click Generate AI Report again — limits reset quickly.
  2. Switch to a different model using the dropdown in the header.
  3. The server retries up to 3 times automatically before surfacing an error.
Model Free RPM Free TPM
Kimi K2 (moonshotai/kimi-k2-instruct) varies varies
Llama 3.3 70B (llama-3.3-70b-versatile) 30 6,000
Qwen3 32B (qwen/qwen3-32b) 30 6,000
GPT-OSS 120B (openai/gpt-oss-120b) varies varies

Check current limits at console.groq.com/settings/limits

🏗️ Architecture

┌──────────────────────────────────────────────────────────┐
│                  React + Vite Frontend                   │
│  ┌──────────┐  ┌──────────┐  ┌─────────┐  ┌──────────┐  │
│  │ Step 1   │→ │ Step 2   │→ │ Step 3  │→ │ Step 4   │  │
│  │ SMILES   │  │ Props +  │  │ HOMO-   │  │ AI Report│  │
│  │ Input    │  │ DFT Data │  │ LUMO    │  │+ExportMenu│  │
│  └────┬─────┘  └────┬─────┘  │ Chart   │  └────┬─────┘  │
│       │             │        └─────────┘       │        │
└───────┼─────────────┼───────────────────────── ┼────────┘
        │             │                           │
   ┌────▼─────────────▼───────────────────────────▼────┐
   │              Netlify Serverless Functions          │
   │  ┌──────────────────────┐  ┌─────────────────────┐ │
   │  │  /analyze            │  │  /report            │ │
   │  │  • PubChem lookup    │  │  • 3× retry         │ │
   │  │  • DFT dataset match │  │  • Groq API         │ │
   │  │  • k-NN prediction   │  │  • System prompt    │ │
   │  └──────────────────────┘  └─────────────────────┘ │
   └────────────────────────────────────────────────────┘

🧪 Quantum Chemistry Knowledge Base

The LLM report generator has PhD-level expertise covering:

  • Wavefunction theory: HF, MP2–MP4, CCSD, CCSD(T) O(N⁷), EOM-CCSD, ADC, CISD, FCI
  • Multi-reference: CASSCF, CASPT2, NEVPT2, MRCI+Q, DMRG, Selected-CI
  • DFT — Jacob's Ladder: LDA → GGA (PBE, BLYP) → meta-GGA (SCAN, r²SCAN) → hybrid (B3LYP, PBE0, M06-2X) → RSH (CAM-B3LYP, ωB97X-D) → double hybrid (B2-PLYP, XYG3)
  • TD-DFT: Casida equation, CT state handling, TDA, excited-state geometry
  • Basis sets: Pople, Dunning cc-pVxZ, def2, Jensen pc-n, ANO-RCC, ECPs
  • Report sections: Identity, structural analysis, electronic properties, recommended methods, dataset comparison, applications, future perspectives

🛠️ Tech Stack

Layer Technology Version
Frontend React + TypeScript 18 / 5.5
Build Vite 5
Styling TailwindCSS 3
Charts Recharts 2
Icons Lucide React 0.468
Markdown react-markdown + remark-gfm 9 / 4
Export jsPDF + docx latest
Backend Netlify Functions (serverless) —
LLM Groq API (free plan) api.groq.com
Models Kimi K2, Llama 3.3 70B, Qwen3 32B, GPT-OSS 120B selectable
Chemistry PubChem PUG-REST API —
Dataset 25 quinones DFT B3LYP/6-31G* —
Hosting Netlify (free tier) —

🚀 Quick Start

git clone https://github.com/Bib569/quinone-task-ai.git
cd quinone-task-ai
npm install

The Groq API key is embedded in the serverless function for the free plan. To use your own key, set it as an environment variable:

GROQ_API_KEY=your_groq_api_key_here

Get a free API key at console.groq.com/keys

Run locally with Netlify CLI (required for serverless functions):

npx netlify dev

Build for production:

npm run build

Deploy to Netlify:

netlify deploy --prod --dir=dist

🔑 Environment Variables

Variable Required Description
GROQ_API_KEY Optional Override the built-in Groq API key. Get one at console.groq.com/keys

The default key is embedded in netlify/functions/report.ts for the free plan. For production with your own key, set it in Netlify dashboard → Site settings → Environment variables and update report.ts to read process.env.GROQ_API_KEY.

📁 Project Structure

quinone-task-ai/
├── index.html
├── netlify.toml                    # Netlify build + functions config
├── package.json
├── tailwind.config.js
├── vite.config.ts
├── tsconfig.json
├── src/
│   ├── main.tsx
│   ├── index.css
│   ├── App.tsx                     # Main 4-step wizard UI + state
│   ├── components/
│   │   ├── ExportMenu.tsx          # MD / PDF / DOCX export dropdown
│   │   └── MethodsLibrary.tsx      # QC methods reference panel (modal)
│   └── lib/
│       ├── exportUtils.ts          # jsPDF + docx export engine
│       ├── quinoneData.ts          # 25-molecule DFT dataset + findQuinone()
│       └── quantumChemKnowledge.ts # Full QC knowledge base
└── netlify/
    └── functions/
        ├── analyze.ts              # PubChem lookup + k-NN HOMO-LUMO prediction
        └── report.ts               # LLM report generator (3× retry, system prompt)

🔬 k-NN HOMO-LUMO Prediction

For molecules not in the DFT dataset, a k-NN model predicts HOMO/LUMO/gap:

Features extracted from SMILES (regex):

  • Aromatic carbons (c)
  • Carbonyl groups (=O, minus sulfone oxygens)
  • Halogens (Cl, Br, F)
  • Amino groups (N)
  • Aromatic nitrogens (n)

Algorithm: k=3 nearest neighbours, normalised Euclidean distance, inverse-distance weighted average of DFT reference values.

Confidence thresholds:

Distance Confidence
< 0.25 High
0.25 – 0.60 Medium
> 0.60 Low

Uncertainty: ±0.3–0.5 eV typical. Displayed as orange bar in the chart with amber info panel.

📊 Quinone Reference Dataset

25 quinone derivatives with DFT B3LYP/6-31G* computed values:

Family Count Color in chart HOMO-LUMO Gap Range
Benzoquinones 11 🔵 Blue 3.72 – 4.27 eV
Naphthoquinones 7 🟣 Purple 3.05 – 3.52 eV
Anthraquinones 7 🟡 Amber 2.78 – 3.12 eV
ML-Predicted — 🟠 Orange varies
Your molecule (DFT) — 🔵 Cyan —

🔧 Design Decisions

  1. Task-oriented UX — Step-by-step wizard guides users through a structured analysis pipeline
  2. Recharts visualization — Interactive bar chart with family-based color coding and tooltip
  3. Groq backend — Netlify Functions keep the API key server-side; OpenAI-compatible endpoint, no CORS issues
  4. Model selector — Dropdown in header right-hand side; auto-disables thinking toggle for non-supporting models
  5. Thinking toggle — Purple brain icon for Qwen3 32B; sends reasoning_effort: 'default' + reasoning_format: 'parsed' to Groq; reasoning returned in message.reasoning field
  6. 3× auto-retry — Transparent server-side retry with 1.5 s / 3.0 s backoff; handles 429/502/503
  7. Embedded dataset — Zero-latency DFT lookups without a database
  8. Multi-format export — jsPDF (Latin-1 sanitised) + docx for offline archival of AI reports

📝 Changelog

Version Date Changes
1.6.1 Mar 2026 Fix thinking mode: replace invalid thinking param with correct Groq reasoning_effort=default + reasoning_format=parsed for Qwen3 32B
1.6.0 Mar 2026 Groq API migration: model selector dropdown (Kimi K2 / Llama 3.3 / Qwen3 / GPT-OSS), thinking mode toggle for Qwen3 32B, 429 rate-limit error handling
1.5.0 Mar 2026 PDF export garbled character fix (sanitiseForPDF Latin-1 sanitisation)
1.4.0 Mar 2026 Server-side 3× retry for 502/503/429; improved error messages
1.3.0 Mar 2026 Multi-format export (MD/PDF/DOCX) replaces JSON export
1.2.0 Mar 2026 QC Methods Library panel; comprehensive knowledge base in report prompt
1.1.0 Mar 2026 k-NN HOMO-LUMO prediction; orange bar in chart; bug fix for empty-name matching
1.0.0 Mar 2026 Initial release with 4-step workflow, PubChem, DFT dataset, AI report

📄 License

MIT © 2026

About

QuantumTask - LLM-powered molecular property analyzer for quinone HOMO-LUMO predictions

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages