Implementation guides

1. Go to chat.openai.com → Explore GPTs → Create
2. In the Configure tab, fill in Name, Description, and Instructions
3. Optionally add Conversation Starters for guided UX
4. Upload Knowledge files (PDFs, docs) for RAG-lite behavior
5. Enable Actions to connect external APIs via OpenAPI schema
6. Set visibility: Only me / Anyone with link / Public
7. Click Save, then share the link with your team

You are [BotName], a specialized assistant for [Company].

Your role:
- Help users with [specific tasks]
- Answer questions about [domain]
- DO NOT discuss [off-topic areas]

Tone: [professional/friendly/concise]

If the user asks something outside your scope, say:
"I'm specialized in [domain]. For other questions, please contact [[email protected]]."

-- Enable required extensions
create extension if not exists vector;
create extension if not exists "uuid-ossp";
create extension if not exists pg_trgm;

-- Documents table
create table documents (
  id uuid primary key default gen_random_uuid(),
  user_id uuid not null references auth.users(id) on delete cascade,
  filename text not null,
  content text not null,
  embedding vector(1536),
  created_at timestamptz default now()
);

-- Hybrid search function (semantic + keyword)
create or replace function hybrid_search(
  query_text text,
  query_embedding vector(1536),
  match_count int,
  p_user_id uuid
) returns table(id uuid, content text, filename text, similarity float)
language sql stable as $$
  select id, content, filename,
    (0.7 * (1 - (embedding <=> query_embedding)) +
     0.3 * ts_rank(to_tsvector('english', content), plainto_tsquery('english', query_text))) as similarity
  from documents
  where user_id = p_user_id
  order by similarity desc
  limit match_count;
$$;

-- RLS: users only see their own documents
alter table documents enable row level security;
create policy "users see own docs" on documents
  for all to authenticated using (auth.uid() = user_id);

{
  "name": "get_weather",
  "description": "Get the current weather for a city",
  "parameters": {
    "type": "object",
    "properties": {
      "city": {
        "type": "string",
        "description": "The city name"
      }
    },
    "required": ["city"]
  }
}

• You need a very consistent tone or format across thousands of responses
• Prompt engineering alone can't achieve the desired behavior
• You have 50+ high-quality input/output examples
• Reducing system prompt tokens matters for cost at scale

{"messages":[{"role":"system","content":"You are a helpful assistant."},{"role":"user","content":"[Example question]"},{"role":"assistant","content":"[Ideal answer in your desired style]"}]}
{"messages":[{"role":"system","content":"You are a helpful assistant."},{"role":"user","content":"[Another question]"},{"role":"assistant","content":"[Another ideal answer]"}]}

Minimum: 10 examples. Recommended: 50–100 for meaningful results. Each line is one JSON object.

• 1Prepare 50+ JSONL examples with system, user, assistant messages
• 2Validate JSONL: each line parseable, messages array present
• 3Upload via OpenAI API: openai.files.create() with purpose='fine-tune'
• 4Start job: openai.fineTuning.jobs.create({ model: 'gpt-4o-mini', training_file: id })
• 5Monitor: openai.fineTuning.jobs.retrieve(jobId)
• 6Use your model: model: 'ft:gpt-4o-mini:org:name:id'
• 7Evaluate against base model on held-out test cases