This model is a LoRA fine-tuned version of Microsoft Phi-2, adapted to act as a startup advisor.
It provides structured, actionable guidance to early-stage founders by learning reasoning patterns from real-world startup case studies.

The model focuses on:

Monetization strategy
Cash burn reduction
Strategic pivots
Niche targeting
Data-driven decision-making

This repository contains only LoRA adapters, not the full base model.

Developed by

Sanjay (independent project)

Model type

Decoder-only causal language model (instruction-tuned via LoRA)

Language(s)

English

Finetuned from model

microsoft/phi-2

License

Apache 2.0 (inherits base model license)

Model Sources

Base model: Microsoft Phi-2
Fine-tuning method: Parameter-Efficient Fine-Tuning (LoRA)
Training framework: Hugging Face Transformers + PEFT

Uses

Direct Use

This model is intended to be used as:

A startup advisory chatbot
A decision-support assistant for early-stage founders
A reasoning-focused LLM for business strategy discussions

Example Use Cases

“My startup is burning cash. How can I reach profitability?”
“Should I target a niche or go mass-market?”
“How can I pivot when my current model is failing?”

Out-of-Scope Use

This model should NOT be used for:

Legal advice
Financial investment advice
Medical advice
Regulatory or compliance decisions

Bias, Risks, and Limitations

The model is trained on a limited number of startup case studies, which may bias it toward patterns common in Indian and SaaS/fintech startups.
It may generate overly optimistic strategies if used without external validation.
It does not have real-time market awareness.
Responses should be treated as advisory insights, not authoritative decisions.

How to Get Started

Load the model (LoRA adapters)

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained(
    "microsoft/phi-2",
    load_in_4bit=True,
    device_map="auto",
    trust_remote_code=True
)

model = PeftModel.from_pretrained(
    base_model,
    "sanjusanjay/phi-2-startup-advisor-lora"
)

tokenizer = AutoTokenizer.from_pretrained(
    "sanjusanjay/phi-2-startup-advisor-lora"
)

model.eval()