This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
#
What I Built
Every independent freelancer—whether a video editor, web developer, or designer—has faced the dreaded client message: "Bhai, ek chota sa change aur kar do free me" (Can you just add these two extra features for free?)
Scope creep and vague WhatsApp agreements cost freelancers thousands of dollars in unpaid labor every year. Most freelancers don't use legal contracts because lawyers are expensive, templates are rigid, and clients often chat informally in mixed languages (like Hinglish—Hindi + English).
I built 🛡️ FreelanceShield OS, an open-weight AI micro-SaaS that turns messy, informal WhatsApp/DM client notes into a complete 4-in-1 Business Protection Suite in 1.5 seconds:
📄 Ironclad Scope-of-Work (SoW) & Micro-Contract: Explicitly defines deliverables, out-of-scope boundaries, revision limits, IP lock, and kill fees. 2. 🧾 Printable Proforma Tax Invoice & Work Order: Auto-calculates upfront deposits, milestone splits, and GST/taxes with a 1-clickPrint / Save as PDF view. 3. 💬 Multi-Channel Client Pitches: Ready-to-send onboarding messages inProfessional English, natural Hinglish, and Formal Email so freelancers can get client sign-off without sounding robotic. 4. 🥊 Live Scope-Creep Pushback Simulator & Red-Flag Audit: Audits client notes for hidden loopholes (with a Deal Safety Score) and generates polite, relationship-friendly pushback replies when a client asks for free out-of-scope work mid-project.
Who would pay for it? Independent freelancers, student creators, and micro-agencies who want instant legal & financial protection without paying $50/month for bloated enterprise contract software.
#
Demo
#
Code
https://github.com/jaivintretiya16-04/freelance-shield-os
#
How I Built It
FreelanceShield OS is built in Python with Streamlit (featuring a custom-engineered dark glassmorphic UI) and powered at its core by Open-Weight AI Models:
Dual Open-Source Inference Engine: 1.
☁️ Groq Open-Weight Cloud: Uses open-weight models likeMeta Llama 3.1 (llama-3.1-8b-instant) ,Mixtral 8x7B , andGemma 2 9B with dynamic model discovery—automatically querying active open models via the API so the app never breaks if a model is deprecated.
2.
🏠 100% Offline Local Inference via Ollama: Built-in native REST integration (http://localhost:11434/api/chat ) allowing freelancers to runllama3.2 locally on their laptop with zero internet connection. #
Structured Multi-Part Generation Pipeline: A single deterministic prompt with strict delimiter parsing (===SPLIT=== ) generates the Contract, Multi-lingual Pitches, and Risk Audit in one sub-2-second inference call. #
Interactive Financial & Legal Guardrail Engine: Dynamic sliders for upfront percentage, tax/GST calculation, revision pricing, and legal toggles (IP retention, late payment penalties, and kill fees) that inject hard mathematical constraints directly into both the LLM prompt and the HTML/CSS Printable Invoice.
#
Why Does Open Innovation Matter?
Building FreelanceShield OS on open-weight models and local inference solved four critical business problems that a closed, proprietary API could not:
🔒 100% Client Confidentiality (Local Ollama Mode): Freelancers deal with sensitive client budgets, NDA terms, and pricing strategies. By supporting local inference viaOllama , a freelancer can generate contracts completely offline on an 8GB RAM laptop—ensuring sensitive client data never touches a third-party server.
2.
🔄 Dynamic Model-Swapping & Zero Vendor Lock-In: While building this project, a specific model endpoint (llama-3.3-70b-versatile ) was deprecated mid-development, and another (qwen ) hit strict output token rate limits. Because our architecture uses open-weight models and dynamic model fetching, we instantly swapped tollama-3.1-8b-instant with zero downtime.
3.
💸 Unit Economics That Actually Work for a Side Hustle: Charging freelancers $3–$5/month for unlimited contracts is impossible if you are paying heavy per-token costs to a closed frontier API. Running open-weight Llama models keeps inference costs near $0.000, giving this side hustle nearly100% gross margins .
4.
🗣️ Nuanced Bilingual (Hinglish) Understanding: Open-weight models like Llama 3 excel at parsing informal Indian-English ("Hinglish") WhatsApp chats and translating them into both strict English legal clauses and culturally natural Hinglish client replies.
#
My First User
I tested FreelanceShield OS with a friend who works as a freelance Instagram Reels video editor. He constantly struggled with clients asking for "just 2 more revisions" or delaying the final 50% payment before taking the video files.
When he pasted his real Hinglish WhatsApp chat into the tool, he was blown away by two things:
- The Printable Invoice that clearly split his ₹6,000 fee into a ₹3,000 upfront deposit and explicitly listed*"₹500 per extra revision after 2 free rounds."*
- The Hinglish WhatsApp Pitch & Pushback Simulator , which gave him a polite message he could actually send to his client without sounding rude or overly formal.
He told me he would gladly pay ₹199–₹299/month (~$3.50/mo) for this tool because saving even one unpaid extra revision pays for the subscription immediately. His next feature request: adding a UPI QR Code generator directly onto the printable invoice so clients can scan and pay the 50% advance instantly!