AI Clone - Your Personal AI Representative

"What do you work on?" "Are you open to new roles?" "What's your tech stack?" Recruiters ask the same questions. Your clone answers them - instantly, accurately, 24/7.

Chat with the live clone →

---

What is an AI Clone?

An AI Clone is a chatbot that represents you. It knows your background, your projects, your skills, and your story - and it answers questions about you the way you would answer them yourself.

You give it your resume, LinkedIn profile, or a simple text file about yourself. It learns from that. Visitors can then have a real conversation with your clone - asking anything from "What projects have you built?" to "Why should I hire you?" - and get accurate, thoughtful answers in your voice.

No hallucinations. No generic responses. Every answer is grounded in what you've actually shared about yourself.

---

Why you need one

For job seekers and professionals:

• Recruiters reach out at all hours. Your clone is always available.
• Share a single link instead of a resume - far more memorable.
• Let your personality and depth come through in a conversation, not a PDF.
• Filter serious interest: someone who chats with your clone is more engaged than someone who skimmed your LinkedIn.

For developers:

• Fork the repo, fill in a .env file, deploy in under 10 minutes.
• Clean, minimal codebase - FastAPI + ChromaDB + Gemini. No magic, no abstractions you can't understand.
• Extend it: swap the LLM provider, add auth, connect a calendar - it's your code.

---

Try it live

Chat with Aayush Gupta's AI clone right now:

https://aayush-ai-clone.fly.dev/

Try asking:

• "What are you currently working on?"
• "What's your experience with AI?"
• "Have you won any hackathons?"
• "Are you open to new opportunities?"
• "How can I reach you?"

---

Create your own clone in 10 minutes

You don't need to be a developer to have a clone. If you can edit a text file and run two commands, you can deploy one.

Step 1 - Fork this repo

Click Fork in the top right on GitHub. You now have your own copy.

Step 2 - Write your knowledge file

Create a file called me.md (plain text works great). Write about yourself:

# About Me
My name is [Your Name]. I'm a [your role] based in [city].

# What I'm working on
Currently building [project] at [company]...

# My skills
[list them out]

# What I'm looking for
[be honest about this]

The more detail you add, the better your clone answers. FAQ format works especially well - write the questions recruiters actually ask, then answer them.

Step 3 - Configure

cp .env.example .env

Edit .env:

GEMINI_API_KEY=AIzaSy...        # free at https://aistudio.google.com/app/apikey
APP_NAME=Your Name
SYSTEM_PROMPT=You are [Your Name]'s AI clone. Speak in first person, be concise and direct.
ADMIN_PASSWORD=choose-a-password

That's it. Four lines.

Step 4 - Deploy to Fly.io

brew install flyctl
fly auth signup
fly apps create your-name-ai-clone
fly volumes create ai_clone_data --size 1 --region ord --app your-name-ai-clone
fly secrets set \
  GEMINI_API_KEY="your-key" \
  APP_NAME="Your Name" \
  ADMIN_PASSWORD="your-password" \
  SYSTEM_PROMPT="your persona instructions" \
  --app your-name-ai-clone
fly deploy

Your clone is live at https://your-name-ai-clone.fly.dev.

Step 5 - Upload your knowledge file

Go to https://your-app.fly.dev/admin, log in with your password, upload your .md or .txt file, and click Upload & re-index. Done.

Share the link. Anyone who visits can chat with your clone.

---

Run locally

git clone https://github.com/your-username/ai-clone.git
cd ai-clone
uv sync
cp .env.example .env   # fill in your values
uv run uvicorn app.main:app --reload

URL	What
http://localhost:8000	Public chat page
http://localhost:8000/admin	Admin dashboard (requires password)
http://localhost:8000/health	Health check

---

How it works

Your file (PDF / TXT / MD)
  → split into chunks
  → each chunk converted to a vector (Gemini Embedding)
  → stored in ChromaDB

Visitor asks a question
  → question converted to vector
  → find most relevant chunks from your file
  → Gemini reads those chunks and generates a grounded answer
  → streamed back token by token

Answers are always grounded in your data. The model is instructed not to use outside knowledge - it won't make things up about you.

---

Stack

Concern	Choice
Backend	FastAPI + Jinja2
Frontend	Server-rendered HTML + vanilla JS (no build step)
Vector store	ChromaDB (local, persistent)
LLM	gemini-2.5-flash with thinking_budget=1024 for accuracy
Embeddings	gemini-embedding-001 via REST
File parsing	pypdf for PDFs, plain read for TXT/MD
Deployment	Fly.io (persistent volume) or Render

---

Project layout

ai-clone/
├── app/
│   ├── main.py              Routes: /, /api/chat, /admin, /admin/ingest, /admin/delete
│   ├── ingest.py            Parse → chunk → embed → store in ChromaDB
│   ├── rag.py               Retrieve top-k chunks → build system prompt
│   ├── providers/
│   │   ├── base.py          Provider protocol
│   │   └── gemini.py        Gemini - embeddings via REST, chat via SDK
│   └── templates/
│       ├── chat.html        Public chat UI (WhatsApp-style, streaming)
│       └── admin.html       Admin dashboard (upload, re-index, config)
├── data/                    Your knowledge files (gitignored)
├── chroma/                  Vector index (gitignored)
├── Dockerfile               For Fly.io deployment
├── fly.toml                 Fly.io config
├── .env.example             Environment variable template
└── pyproject.toml

---

Environment variables

Variable	Required	Description
GEMINI_API_KEY	Yes	Free at aistudio.google.com
APP_NAME	Yes	Your name - shown in the chat header and avatar
SYSTEM_PROMPT	No	Persona and tone instructions for the model
ADMIN_PASSWORD	Yes	Password for the /admin dashboard
DATA_DIR	No	File storage path (default: data/)
CHROMA_DIR	No	Vector index path (default: chroma/)

---

Knowledge file tips

• Plain text beats PDFs - PDF extraction can be noisy. A well-written .md or .txt file gives the model cleaner context and better answers.
• FAQ format works best - write the questions recruiters actually ask, then answer them directly. The model retrieves those Q&A pairs at query time.
• Be specific - dates, company names, project details, numbers. Vague input produces vague output.
• Write in first person - "I built...", "I'm looking for...". The clone speaks in your voice.
• Multiple files are fine - they're all indexed together into one collection.

---

Adding a new LLM provider

The provider interface is simple:

def embed(texts: list[str]) -> list[list[float]]: ...
def embed_query(text: str) -> list[float]: ...
def chat_stream(system: str, history: list[dict], user: str) -> Iterator[str]: ...

Create app/providers/openai.py (or any provider), implement those three methods, swap it in main.py. The rest of the app doesn't change.

---

Diagnostic

If you get embedding errors, run this to see which Gemini models are available for your API key:

uv run python check_api.py AIzaSy...your_key_here