adamabuhamdan/startup-advisor-dataset

{
  "messages": [
    {
      "role": "system",
      "content": "You are an expert startup and business advisor. Provide concise, actionable advice grounded in proven frameworks and real-world examples from leading companies."
    },
    {
      "role": "user",
      "content": "I'm at the **Scaling** stage of building my startup.\nGive me actionable advice on **blitzscaling speed**.\nRelevant themes: speed, growth, market_capture.\nComplexity level: advanced."
    },
    {
      "role": "assistant",
      "content": "Prioritize speed over efficiency when your market window is closing. Accept chaos, technical debt, and management strain as temporary costs of capturing a winner-take-all position before rivals do.\n\n**Why this matters:** In fast-moving markets, the first company to scale dominates distribution, talent, and network effects — making catch-up nearly impossible.\n\n**Real-world example (2025):** OpenAI blitzscaled its API platform in 2025, accepting early instability to lock in developers before Google and Anthropic consolidated the market.\n\n**Avoid:** `optimizing_too_early`, `waiting_for_perfect_infrastructure`, `prioritizing_margins_over_growth`"
    }
  ],
  "_meta": {
    "advice_id": "advice_scaling_001",
    "source_file": "Blitzscaling.md",
    "category": "Scaling",
    "stage": "Scaling",
    "topic": "blitzscaling_speed",
    "complexity": "advanced",
    "tags": ["speed", "growth", "market_capture"]
  }
}

from datasets import load_dataset

ds = load_dataset("adamabuhamdan/startup-advisor-dataset", split="train")
print(ds[0])

from trl import SFTTrainer, SFTConfig
from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "Qwen"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model     = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto")

dataset = load_dataset("adamabuhamdan/startup-advisor-dataset", split="train")

trainer = SFTTrainer(
    model=model,
    tokenizer=tokenizer,
    train_dataset=dataset,
    args=SFTConfig(
        output_dir="./startup-advisor-ft",
        num_train_epochs=3,
        per_device_train_batch_size=2,
        gradient_accumulation_steps=8,
        learning_rate=2e-4,
        bf16=True,
    ),
)
trainer.train()

ds = load_dataset("adamabuhamdan/startup-advisor-dataset", split="train")

# Keep only advanced Scaling entries
import json
advanced_scaling = [
    row for row in ds
    if row["_meta"]["category"] == "Scaling"
    and row["_meta"]["complexity"] == "advanced"
]

@dataset{abuhamdan2025startup,
  author    = {Adam Abu Hamdan},
  title     = {Startup Advisor Dataset},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/adamabuhamdan/startup-advisor-dataset}
}