Privacy-preserving proof-of-authorship with AI fingerprinting and Midnight blockchain receipts. Built for the Midnight × MLH Hackathon 2026.

Live Demo: https://ghostwrite-midnight-hackathon.vercel.app

Content theft is rampant. Ghostwriters, researchers, and journalists have their work paraphrased and claimed by others — and have no way to prove they wrote it first without exposing the original text.

Existing tools like Turnitin or Copyscape require you to submit your original writing, destroying any expectation of privacy.

GhostWrite lets you:

1. Write or generate a document using AI
2. Seal it — a cryptographic hash is stored on the Midnight Preprod blockchain, timestamping your authorship forever
3. Prove derivation — if someone paraphrases your work, paste their text + your receipt hash and our AI fingerprinting engine computes a similarity score without ever seeing your original

Your content stays private. Only the proof is public.

When you secure a document, GhostWrite silently:

1. Sends your text to Groq's Llama-3.3-70B model
2. Extracts a semantic fingerprint: topics, entities, tone, vocabulary richness, sentence cadence, transition frequency
3. Stores the fingerprint in Supabase, linked to your Midnight receipt hash

When someone checks authorship:

1. Their suspicious text is fingerprinted the same way
2. A weighted similarity algorithm compares the two fingerprints (no original text is ever used)
3. Returns a similarity score with reasoning

Scoring weights: Topics (50%) · Style (20%) · Tone (15%) · Entities (15%)

User → Next.js App → Groq LLM (fingerprint extraction)
                   → Midnight Blockchain (hash receipt)
                   → Supabase (fingerprint + document storage)

Verifier → /api/authorship → Supabase (fetch stored fingerprint)
                           → Groq LLM (extract suspicious text fingerprint)
                           → semanticCompare() → similarity score

• Node.js 18+
• A Supabase project
• A Groq API key (free)

git clone https://github.com/Saurabhkohli8476/ghostwrite-midnight-hackathon.git
cd ghostwrite-midnight-hackathon
npm install

Create .env.local in the project root:

# Supabase
NEXT_PUBLIC_SUPABASE_URL=your_supabase_project_url
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key

# Groq (free at console.groq.com)
GROQ_API_KEY=your_groq_api_key

Run this SQL in your Supabase SQL Editor:

-- Documents table
create table cover_letters (
  id uuid primary key default gen_random_uuid(),
  user_id uuid references auth.users not null,
  job_title text,
  company text,
  job_description text,
  user_experience text,
  generated_letter text,
  midnight_hash text,
  midnight_tx text,
  is_secured boolean default false,
  secured_at timestamptz,
  created_at timestamptz default now()
);

-- Fingerprints table
create table document_fingerprints (
  id uuid primary key default gen_random_uuid(),
  document_id uuid references cover_letters(id) on delete cascade,
  fingerprint jsonb not null,
  created_at timestamptz default now()
);

-- Row Level Security
alter table cover_letters enable row level security;
alter table document_fingerprints enable row level security;

create policy "Users manage own documents" on cover_letters
  for all using (auth.uid() = user_id);

create policy "Users manage own fingerprints" on document_fingerprints
  for all using (
    auth.uid() = (select user_id from cover_letters where id = document_id)
  );

npm run dev
# Open http://localhost:3000

1. Sign up with any email
2. Go to New Document → paste a job description → skip to Step 3
3. Paste any text as your "draft" → click Seal on Midnight
4. Wait for success toast → go to Archive → copy the document hash
5. Go to Verify → Check Authorship
6. Paste a paraphrased version of your text + the hash → click Analyze
7. See the similarity score and reasoning

1. Saurabh Kohli
2. Robin Rajratn
3. Deepika Kohli

Built at the Midnight × MLH Hackathon 2026.