A newer version of the Streamlit SDK is available: 1.57.0
metadata
title: PakFreelance
emoji: 🚀
colorFrom: purple
colorTo: blue
sdk: streamlit
app_file: app.py
pinned: false
🇵🇰 PakFreelance — AI Freelance Toolkit
Professional AI Agent Toolkit for Pakistani Freelancers
AMD Developer Hackathon 2026 | Track 1: AI Agents & Agentic Workflows
Powered by AMD MI300X through Groq's optimized API, running Llama 3.3 70B for fast inference. Deployed on HuggingFace Spaces.
🎯 Overview
PakFreelance is an AI-powered toolkit built with CrewAI agents that helps freelancers across Pakistan and South Asia:
✅ Generate winning proposals in seconds (5-agent AI orchestration)
✅ Detect client scams before wasting time (red flag analysis)
✅ Calculate optimal rates based on skills and goals
✅ Score proposals with detailed improvement feedback
✅ Write professional bios for Upwork/Fiverr
✅ Analyze client websites for legitimacy
✅ One-click Power Mode for complete job analysis
Built for the 4M+ freelancers in Pakistan and South Asia using AMD MI300X infrastructure, Llama 3.3 70B, and CrewAI.
🚀 Key Features
⚡ Power Mode (One-Click Analysis)
Paste a job posting → Get risk assessment + proposal + pricing strategy in 3 columns
✈️ Proposal Generator
5 specialized AI agents working together:
- Scout - Extract job requirements
- Researcher - Analyze client context
- Writer - Generate personalized proposal (150-200 words)
- Pricer - Recommend optimal pricing
- Reviewer - Polish & finalize
🚩 Red Flag Detector
- Identifies scam patterns
- Detects unrealistic budgets
- Spots exposure-only offers
- Provides safe response templates
⭐ Proposal Scorer
- Scores out of 100
- Breakdown by 7 criteria (Hook, Credibility, Understanding, Solution, etc.)
- Top 3 specific improvement suggestions
💰 Rate Calculator
- Pakistan-specific pricing (30-50% lower than US rates)
- Experience multiplier
- Monthly goal calculation
- Tax & platform fee buffer
- Three-tier pricing strategy (Start/Target/Premium)
📝 Profile Bio Writer
- Upwork/Fiverr optimized
- Compelling headlines
- Professional 150-200 word bios
- Personality-driven tone
🌐 Website Analyzer
- Extract website content
- Analyze for red flags
- Assess client legitimacy
📄 Document Reader
- Upload CV & Job PDFs
- Extract relevant info
- Provide smarter analysis
🛠️ Tech Stack
| Component | Technology | Purpose |
|---|---|---|
| AI Framework | CrewAI | Multi-agent orchestration |
| LLM | Llama 3.3 70B (via Groq) | Fast inference on AMD MI300X |
| Infrastructure | AMD MI300X (via Groq API) | GPU acceleration |
| Web Framework | Streamlit | Beautiful UI |
| Document Export | python-docx | Word document generation |
| PDF Support | PyPDF2 | PDF reading |
| Web Scraping | requests | Website text extraction |
📋 System Architecture
┌─────────────────────────────────────────────────────────┐
│ PAKFREELANCE APP │
│ (Streamlit Web Interface) │
└──────────────────┬──────────────────────────────────────┘
│
┌──────────┴──────────┐
│ │
▼ ▼
┌────────────┐ ┌──────────────┐
│ Job Input │ │ User Profile │
└──────┬─────┘ └──────┬───────┘
│ │
└───────┬───────────┘
▼
┌──────────────────────┐
│ CrewAI Agents: │
├──────────────────────┤
│ • Scout Agent │
│ • Researcher Agent │
│ • Writer Agent │
│ • Pricer Agent │
│ • Reviewer Agent │
└───────┬──────────────┘
│
▼
┌──────────────────────┐
│ Groq API Call │
│ (Llama 3.3 70B) │
│ │
│ AMD MI300X Powered │
└───────┬──────────────┘
│
▼
┌──────────────────────┐
│ Output Results: │
├──────────────────────┤
│ • Proposals │
│ • Red Flags │
│ • Rates │
│ • Scores │
│ • Bios │
└──────────────────────┘
📦 Installation
1. Clone Repository
git clone https://github.com/yourusername/pakfreelance.git
cd pakfreelance
2. Create Virtual Environment (Optional but Recommended)
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
3. Install Dependencies
pip install -r requirements.txt --break-system-packages
4. Set Up Environment Variables
Create .env file in project root:
GROQ_API_KEY=your_groq_api_key_here
Get API Keys:
- Groq: https://console.groq.com (Free tier: 12,000 TPM)
5. Run Application
streamlit run app.py
Visit: http://localhost:8501
🎮 Usage Guide
Power Mode (Quickest)
- Click ⚡ Power Mode in sidebar
- Paste job posting
- Click 🚀 Run Complete Analysis
- Get risk assessment + proposal + rates in 3 columns
- Click expanders for full results
Proposal Generator
- Paste job posting
- (Optional) Enter your skill & experience
- Click ✈️ Generate Proposal
- Review proposal
- Download as Word document (.docx)
Red Flag Detector
- Choose: Job Description or Website URL
- Paste job or enter website
- Click 🔍 Analyze for Red Flags
- Get risk level + specific red flags
- Copy safe response template if HIGH RISK
Rate Calculator
- Enter skill (e.g., "Python Developer")
- Enter years of experience
- Enter monthly income goal (USD)
- Enter available hours/week
- Click 💰 Calculate Rate
- Get optimal hourly rate + pricing strategy
Proposal Scorer
- Paste your proposal
- Click ⭐ Score My Proposal
- Get score out of 100 + breakdown + improvements
Profile Bio Writer
- Enter name, skill, experience
- (Optional) Specialization & achievements
- Click 📝 Generate Bio
- Get headline + professional bio
📊 Example Output
Proposal Generated
Professional Proposal
Generated by: PakFreelance AI Toolkit
Building a complete e-commerce platform with a MERN stack is exactly
the kind of challenge I enjoy solving. I have 5+ years of experience
building scalable applications for e-commerce clients, including 10+
projects with React, Node.js, and MongoDB.
I'll design a responsive UI using React, implement a robust backend
with Node.js and MongoDB, integrate Stripe for payments, and set up
JWT authentication. You'll get a GitHub repo and full documentation.
I'm available immediately and can deliver within your 20-day timeline...
Best regards,
[Your Name]
---
Generated by PakFreelance AI
Powered by AMD MI300X • CrewAI • Llama 3.3 70B
Rate Calculation
OPTIMAL HOURLY RATE: $15 - $30/hour
CALCULATION BREAKDOWN:
- Monthly goal: $3000
- Weeks per month: 4.3
- Hours per week: 30
- Base rate needed: $22.08/hour
- Market rate for Prompt Engineering: $25 - $50/hour
- Experience multiplier (2 years): 1.2
- Adjusted rate: $20 - $40/hour
- Considering Pakistani context: $15 - $30/hour
PRICING STRATEGY:
- Start at: $15/hour (to build reviews)
- Target rate: $25/hour (once established)
- Premium rate: $35/hour (for repeat clients)
🏗️ Project Structure
pakfreelance/
├── app.py # Main Streamlit app (650+ lines)
├── agents.py # CrewAI agents + tasks (150+ lines)
├── red_flags.py # Red flag detection agent
├── proposal_scorer.py # Proposal scoring agent
├── rate_calculator.py # Rate calculation agent
├── profile_bio_writer.py # Bio generation agent
├── requirements.txt # Python dependencies
├── .env.example # Example environment variables
├── README.md # This file
└── .gitignore # Git ignore patterns
📋 Dependencies
crewai==0.25.0 # Multi-agent orchestration
streamlit==1.28.0 # Web UI framework
groq==0.4.1 # Groq API client
litellm==1.0.0 # LLM wrapper
python-dotenv==1.0.0 # Environment variables
python-docx==0.8.11 # Word document creation
PyPDF2==3.0.1 # PDF reading
requests==2.31.0 # HTTP requests
duckduckgo-search==3.8.4 # Web search (optional)
Full list: See requirements.txt
🎯 AMD Technology Integration
Infrastructure Overview
This project leverages AMD MI300X GPUs through Groq's optimized API:
- LLM: Llama 3.3 70B (AMD-optimized model)
- Inference: Groq API (runs on AMD MI300X)
- Performance: Fast token generation (100+ tokens/sec)
- Hosting: HuggingFace Spaces (production-ready)
How AMD Powers This Application
AMD MI300X GPU Compute
- All AI inference runs on AMD Instinct MI300X architecture
- Powered via Groq's AMD-optimized endpoints
- Fast token generation (100+ tokens/sec)
- Cost-effective for production workloads
ROCm Support
- Llama 3.3 70B is ROCm-compatible
- Optimized for AMD Instinct MI300X
- Can scale to local AMD GPU clusters with ROCm
CrewAI Multi-Agent Orchestration on AMD
- 5 specialized agents working in sequence
- All processing runs on AMD MI300X hardware
- Demonstrates scalable agent architecture
Why AMD for This Project
- ✅ Cost-effective inference (free API tier)
- ✅ Production-ready infrastructure
- ✅ Supports open-source models (Llama 3.3)
- ✅ Scalable to enterprise deployments
- ✅ No vendor lock-in (ROCm is open source)
🚀 Deployment
Hugging Face Spaces (Recommended)
- Create Hugging Face Space at https://huggingface.co/spaces
- Select Streamlit SDK
- Connect GitHub repository
- Add environment variable:
GROQ_API_KEY - Deploy automatically (3-5 minutes)
- Get live URL for submission
Local Testing
streamlit run app.py
🐛 Troubleshooting
"RateLimitError from Groq"
Wait 8 seconds between requests. Free tier: 12,000 TPM (tokens per minute)
"Invalid API Key"
Check .env file - make sure key starts with gsk_ and has no extra spaces
"ModuleNotFoundError"
pip install module-name --break-system-packages
📝 License
MIT License - See LICENSE file for details
🙏 Acknowledgments
- AMD Developer Program - Infrastructure & credits
- lablab.ai - Hackathon platform
- CrewAI - Multi-agent orchestration framework
- Groq - Fast LLM inference on AMD MI300X
- Llama 3.3 70B - Powerful open-source language model
- Streamlit - Beautiful web framework
📞 Support
- Documentation: See this README
- Issues: Open a GitHub issue
- Setup Help: Check
.env.examplefor environment variables
🚀 Future Roadmap
Current (v1.0)
- ✅ Multi-agent proposal generation
- ✅ Red flag detection
- ✅ Rate calculation
- ✅ Proposal scoring
- ✅ Bio generation
- ✅ Website analysis
- ✅ Power Mode
Planned (v2.0)
- Document upload & analysis
- Browser extension
- Mobile app
- Multi-language support
Stretch Goals (v3.0)
- Fine-tuned models
- Real-time job scraping
- Auto-apply features
- Client management dashboard
Built with ❤️ for Pakistani Freelancers
Powered by AMD MI300X through Groq
AMD Developer Hackathon 2026 - Track 1: AI Agents
Last Updated: May 8, 2026