Bridgr

Serious talent. No noise. — A verified freelance marketplace built for students and early-career developers.

Comment

Inspiration

Freelance marketplaces are broken for early-career developers.

Junior engineers, CS students, and open-source contributors routinely get passed over because their résumés look thin—even when their GitHub commit histories, competitive programming ratings, and portfolio work tell a completely different story. Meanwhile, small startups and indie founders waste hours sifting through generic proposals with no reliable way to verify whether a developer can actually deliver.

We built Bridgr to bridge that gap. The name says it all: we connect both sides of this trust gap with verifiable proof instead of self-written claims.

What it does

Bridgr is a full-stack freelance marketplace purpose-built for early-career developers. It has two distinct sides:

For clients (startups, indie founders, and small companies):

  • Post a project brief — describe what you need, the required skills, budget, and timeline.
  • AI Scope Analyzer — can't clearly articulate your requirements? Describe your idea in a sentence, and Bridgr's NVIDIA NIM-powered AI generates a complete, structured brief for you.
  • Smart matching — Bridgr scores every available freelancer against your brief using a weighted algorithm (62% skill overlap, 22% budget compatibility, and 16% verification strength) and surfaces the top-ranked candidates.
  • Transparent score breakdown — see exactly why a candidate ranked highly, with their skill match ratio, budget fit, and verification score clearly explained.
  • Milestone-based contracts — no hourly guessing. Fund work in stages, with each milestone held in escrow and released only after you approve the deliverable.
  • AI Milestone Planner — post a brief, and the AI (powered by Meta Llama 3.3 70B via NVIDIA NIM) generates a sequenced milestone plan that matches your budget exactly.
  • Integrated chat for every brief and contract, with an AI Agreement Summarizer that reads the conversation and drafts a plain-English agreement with suggested milestones.
  • Wallet dashboard — view transaction history, monthly spending charts, escrow status, and payout management.
  • Dispute resolution flow for disagreements over milestone approvals.

For freelancers (college students, fresh graduates, and open-source contributors):

  • Verified identity onboarding — connect GitHub, LeetCode, and Codeforces. We verify real public activity, including commit counts, problem-solving statistics, and competitive programming rankings—not just profile links.
  • Profile strength meter — see a live score as you complete your profile, so you know exactly what to improve to rank higher in matches.
  • AI Pitch Composer — once matched to a project, the AI drafts a personalized proposal based on your verified skills and the project's requirements, with selectable tones (Professional, Friendly, or Confident).
  • Work log — log hours for each milestone, and let clients approve or reject entries directly.
  • Earnings wallet — track total earnings, funds held in escrow, pending releases, and platform fees.
  • Request payouts directly from the dashboard.

How we built it

Layer Technology
Framework Next.js 16 (App Router), React 19, TypeScript
UI shadcn/ui, Tailwind CSS v4, Lucide React
AI NVIDIA NIM API with Meta Llama 3.3 70B Instruct for scope analysis, milestone planning, pitch generation, and agreement summarization
Database PostgreSQL on Amazon Aurora (using AWS RDS Signer for IAM authentication)
Payments Stripe (milestone escrow checkout flow with a 6% platform fee)
Auth Custom session-based authentication with server-only cookie access
Data Fetching SWR for real-time polling in the chat hub (4-second refresh interval)
Deployment Vercel with Vercel Analytics
Prototyping v0 by Vercel for rapid UI scaffolding

The matching algorithm is a pure TypeScript scoring function in lib/matching.ts:

[ \text{Score} = 0.62 \cdot \text{SkillOverlap} + 0.22 \cdot \text{BudgetCompatibility} + 0.16 \cdot \text{VerificationStrength} ]

Each sub-score is calculated as a ratio (0–1), multiplied by its weight, and scaled to a score between 0 and 100. Budget compatibility awards partial credit for near matches rather than using a strict binary cutoff.

All AI requests are sent to the NVIDIA NIM OpenAI-compatible endpoint on the server, ensuring the API key is never exposed to the client. Every endpoint also includes a deterministic fallback response in case the AI service is unavailable.

Challenges we ran into

  • Verification UX vs. backend reality — presenting GitHub, LeetCode, and Codeforces data in a way that feels trustworthy rather than merely decorative required carefully designing the onboarding flow and score breakdown UI so users understand what was verified and why it matters.
  • AI reliability — LLMs don't always return valid JSON or respect budget constraints exactly. We implemented strict server-side parsing, budget validation, and automatic fallback templates to ensure the application continues working even when the model behaves unexpectedly.
  • Escrow state machine — milestones move through six states (pending → funded → submitted → approved → paid → disputed). Keeping the UI, database, and payment logic synchronized across role-specific dashboards required careful state management.
  • Real-time experience without WebSockets — we used SWR's refreshInterval to provide a real-time feel while remaining compatible with a serverless deployment model.
  • Dual-role dashboards — clients and freelancers see different views of the same contracts, wallets, and projects. Managing role-based routing and shared state cleanly was an ongoing design challenge.

Accomplishments that we're proud of

  • A fully integrated milestone escrow flow that allows clients to fund milestones, monitor escrow, approve work, and release payments within a single workflow.
  • Three AI-powered features backed by the NVIDIA NIM API, each with graceful fallbacks: Scope Analyzer, Milestone Planner, and Pitch Composer/Agreement Summarizer.
  • A transparent matching system that explains exactly how each freelancer was scored by breaking down skill overlap, budget compatibility, and verification strength—something traditional freelance marketplaces rarely provide.
  • A verification-first onboarding flow that emphasizes real developer activity rather than relying solely on self-declared skills.
  • A complete work log system where freelancers record milestone progress and clients approve or reject submissions, creating a clear audit trail for every contract.

What we learned

  • NVIDIA NIM's OpenAI-compatible API made it straightforward to integrate powerful open models such as Llama 3.3 70B without significantly changing our application's request patterns. The bigger challenge was prompt engineering and defensive response parsing.
  • Designing for two different user roles in a single application requires careful thinking about permissions, data ownership, and trust.
  • Milestone-based payment workflows are more complex than they first appear. A clear state machine is essential to keep transitions predictable and secure.
  • Graceful degradation proved invaluable. Building deterministic fallback responses for AI endpoints made the product significantly more reliable and demo-friendly.

What's next for Bridgr

  • Live GitHub, LeetCode, and Codeforces integrations using OAuth and official APIs to replace seeded verification data.
  • Stripe Connect payouts for direct freelancer payments.
  • Ratings and reviews so reputation grows after every completed project.
  • AI-assisted dispute resolution that analyzes conversations and work logs to provide neutral recommendations.
  • Email and in-app notifications for milestone updates, new messages, and matching events.
  • Public freelancer profiles that allow clients to discover talent outside of AI-generated recommendations.

Built With

  • amazon-aurora-postgresql
  • aws-iam-authentication
  • aws-rds
  • lucide-react
  • meta-llama-3.3-70b
  • next.js
  • nvidia-nim-api
  • pnpm
  • postgresql
  • react
  • shadcn/ui
  • stripe
  • swr
  • tailwind-css
  • typescript
  • v0
  • vercel
  • vercel-analytics
Share this project:

Updates