ProposalKaro

An intelligent 5-stage Gemini multi-agent swarm that turns raw freelance job postings into highly competitive, non-generic proposals to help freelancers win more contracts.

Comment

Inspiration

Freelancers today face a massive wave of competition from generic, AI-generated proposals. Clients are flooded with low-quality, copy-pasted ChatGPT responses within minutes of posting a job. We wanted to shift the question away from "Can AI write a proposal?" to "Can AI strategize a winning bid?" ProposalKaro was inspired by the need to give freelancers an enterprise-grade AI edge—focusing on strategy, market positioning, and direct-response psychology rather than just text generation.

What it does

ProposalKaro is a multi-agent AI swarm that turns a raw freelance job description into a highly competitive, non-generic proposal. Instead of using a single zero-shot prompt, it routes the task through 5 specialized Gemini reasoning stages:

  1. Job Intelligence: Extracts mandatory requirements and hidden urgency signals.
  2. Positioning Strategist: Maps the freelancer's specific skills against client pain points.
  3. Market Pricer: Estimates a competitive bid strategy based on effort.
  4. Proposal Writer: Drafts a concise, fluff-free direct response.
  5. Quality Control Critic: Gemini evaluates the generated proposal for "robotic AI tone" and missing requirements before surface rendering.

How we built it

We utilized a Next.js 14 frontend with a sleek, premium dark-mode glassmorphism aesthetic built on Tailwind CSS. The heavy lifting is done by a Python FastAPI backend powered by the google.generativeai SDK, leveraging the lightning-fast inference of Gemini 3.6-flash.

To handle hackathon quota limits (2 RPM) while retaining our complex multi-agent logic, we compiled our sequential reasoning tasks into a massive chained pipeline that executes the entire 5-stage strategy in a single ~15-second lightning-fast burst.

For our SaaS business model, we implemented a zero-database mechanism. Earning unlimited "Swarm Pro" access requires a native on-chain transaction via the Solana Phantom Wallet. Our backend verifies the RPC ledger in real-time, instantly syncing Pro-entitlement across devices without needing user accounts or passwords.

Challenges we ran into

Integrating multiple autonomous agents sequentially posed significant latency and timeout challenges. Hitting multiple endpoints consecutively caused 60-second frontend network aborts and 429 quota exhaustion. We overcame this by heavily optimizing our prompt structure for Gemini's 3.6-flash model, consolidating the logic funnel to deeply chain the thought process without hitting multiple heavy network hops.

Accomplishments that we're proud of

  • Successfully engineering a self-evaluating Quality Score loop where Gemini actively grades and critiques its own generated output before completion.
  • Building a Vercel-like SaaS aesthetic that feels incredibly premium, elevating it beyond a typical hackathon prototype.
  • Integrating Web3 Phantom Wallet endpoints for instantaneous on-chain SaaS micro-subscriptions.

What's next for ProposalKaro

We plan to introduce "Client Counter-Offer Simulation" (where Gemini predicts how a client might reply to your proposal) and launch our native Chrome Extension to overlay Swarm Intelligence directly on top of Upwork and Fiverr job feeds!

Built With

Share this project:

Updates