Ever wanted to ask a researcher about their work without scheduling a meeting?
This is an AI clone of Prof. Milan Amrut Joshi — trained on all 15+ published research papers, book chapters, and conference publications. Ask it anything about Topological Data Analysis, Multi-Objective Optimization, Machine Learning, or Functional Analysis.
┌─────────────────────────────────────────────────────────────────┐
│ ASK ME ANYTHING ABOUT │
├─────────────────┬───────────────────┬───────────────────────────┤
│ 📐 TDA │ ⚙️ Optimization │ 🤖 Machine Learning │
│ │ │ │
│ • Persistent │ • NSGA-II vs │ • SVM with noisy data │
│ homology │ MOALO vs MODA │ • PSO hyperparameter │
│ • Betti │ • Response │ tuning │
│ numbers │ surface │ • Ensemble methods │
│ • Simplicial │ modeling │ • Feature extraction │
│ complexes │ • Pareto │ • Breast cancer │
│ • Feature │ optimal │ classification │
│ extraction │ solutions │ • Human activity │
│ • Healthcare │ • Composite │ recognition │
│ big data │ plate design │ • Metamodeling │
├─────────────────┼───────────────────┼───────────────────────────┤
│ 🧮 Mathematics │ 🔬 Engineering │ 📚 General Research │
│ │ │ │
│ • Almost │ • Micro- │ • "Explain your best │
│ periodic │ machining │ paper award" │
│ functions │ • Turbulent │ • "What is TDA?" │
│ • Banach │ flows │ • "Compare DTM vs DJM" │
│ algebras │ • Fatigue │ • "Summarize your │
│ • Fixed point │ strength │ Springer chapters" │
│ theorems │ • Plasma PDEs │ • "Research timeline?" │
│ • Compact │ • Steel │ • "What should I cite?" │
│ operators │ prediction │ │
└─────────────────┴───────────────────┴───────────────────────────┘
🔬 "What is persistent homology and how did you use it?"
AI Clone: Persistent homology is a central tool in Topological Data Analysis that captures the "shape" of data across multiple scales. In my work, I've applied it in several ways:
Feature Extraction (ICIVC-2021, Best Paper Award): I developed a novel framework that uses persistent homology to generate topological features — like persistence diagrams and Betti curves — as input features for downstream ML models. These topological features capture shape-based patterns that traditional statistical features miss.
Healthcare Big Data (IJAER, 2019): I surveyed how TDA methods including persistent homology can extract clinically meaningful patterns from high-dimensional medical datasets.
Fatigue Strength Prediction (IOP, 2020): I showed that topologically-informed features outperform standard statistical methods when predicting steel fatigue strength.
📄 Key Papers: #5, #6, #8, #14, #15 in my publication list
⚙️ "Which optimization algorithm should I use for multi-objective problems?"
AI Clone: Based on my comparative study in Materials (2021), here's my guidance:
- NSGA-II: Best overall performer — robust, well-documented, excellent for 2-3 objective problems. Use this as your default choice.
- MOALO: Good for problems with complex, non-convex Pareto fronts. The ant lion metaphor provides unique exploration dynamics.
- MODA: Effective for certain problem structures but less consistent across different benchmark functions.
For engineering optimization specifically (like composite plate design), combining response surface modeling with multi-objective optimization (as we showed in Steel & Composite Structures, 2019) can dramatically reduce computational cost while maintaining solution quality.
📄 Key Papers: #1, #2
🧮 "Explain your work on almost periodic functions"
AI Clone: My earliest work (2010-2011) was in pure functional analysis:
In my 2010 paper with Prof. Padhye, we proved that several seemingly different definitions of almost periodic functions on locally compact groups are actually equivalent. This is important because different authors use different definitions, and our work showed they all describe the same mathematical object.
In 2011, I extended this to characterize collectively compact sets of operators on the Banach algebra of almost periodic functions. This bridges operator theory with harmonic analysis.
These pure mathematical foundations later informed my applied work — the topological thinking from functional analysis directly influenced my approach to TDA and persistent homology.
📄 Key Papers: #9, #10
┌──────────────────────────────────────────────────────────────┐
│ ASK MY AI CLONE │
│ │
│ ┌──────────┐ ┌──────────────┐ ┌──────────────────┐ │
│ │ User │───>│ Streamlit │───>│ Query Engine │ │
│ │ Question │ │ Frontend │ │ │ │
│ └──────────┘ └──────────────┘ │ 1. Embed query │ │
│ │ 2. Vector search │ │
│ ┌──────────────────────────────┐ │ 3. Retrieve top │ │
│ │ Knowledge Base │ │ chunks │ │
│ │ ┌────────────────────────┐ │ │ 4. Build prompt │ │
│ │ │ 15 Research Papers │ │<───│ 5. Call GPT-4o │ │
│ │ │ (PDF → Chunks → Embed) │ │ │ 6. Stream answer │ │
│ │ └────────────────────────┘ │ └──────────────────┘ │
│ │ ┌────────────────────────┐ │ │
│ │ │ ChromaDB / FAISS │ │ ┌──────────────────┐ │
│ │ │ Vector Store │ │ │ Response with │ │
│ │ └────────────────────────┘ │ │ 📄 Citations │ │
│ └──────────────────────────────┘ │ 📊 Confidence │ │
│ │ 🔗 Paper Links │ │
│ └──────────────────┘ │
└──────────────────────────────────────────────────────────────┘
| Component | Technology |
|---|---|
| Frontend | Streamlit |
| Embeddings | OpenAI text-embedding-3-small |
| Vector Store | ChromaDB (local) / FAISS |
| LLM | GPT-4o via OpenAI API |
| PDF Processing | PyMuPDF + LangChain |
| Framework | LangChain + LlamaIndex |
| Hosting | Hugging Face Spaces / Streamlit Cloud |
git clone https://github.com/mlnjsh/ask-my-ai-clone.git
cd ask-my-ai-clone
pip install -r requirements.txtcp .env.example .env
# Edit .env and add your OPENAI_API_KEYpython scripts/index_papers.pystreamlit run app/main.pyVisit http://localhost:8501 and start chatting! 🎉
| # | Paper | Year | Topics |
|---|---|---|---|
| 1 | Comparison of NSGA-II, MOALO and MODA | 2021 | Multi-Objective Optimization, Micro-Machining |
| 2 | Response Surface for Composite Plates | 2019 | Response Surface, Structural Optimization |
| 3 | PSO-Tuned SVM for Turbulent Flows | 2020 | PSO, SVM, Metamodeling, CFD |
| 4 | SVM with Imperfect Training Data | 2014 | SVM Robustness, Noisy Data |
| 5 | TDA for Healthcare Big Data | 2019 | TDA, Healthcare, Persistent Homology |
| 6 | Persistent Homology Survey (Springer) | 2020 | Persistent Homology, Machine Intelligence |
| 7 | Hamiltonian PDEs in Plasma (Springer) | 2020 | Hamiltonian Systems, Plasma, PDEs |
| 8 | TDA Feature Selection for Steel | 2020 | TDA, Fatigue Strength, Feature Selection |
| 9 | Almost Periodic Functions on Groups | 2010 | Harmonic Analysis, Locally Compact Groups |
| 10 | Compact Operators on Banach Algebra | 2011 | Operator Theory, Banach Algebras |
| 11 | Fixed Point Theorems | 2017 | Fixed Point Theory, Best Approximation |
| 12 | DTM vs DJM for Differential Equations | 2017 | Numerical Methods, Functional DEs |
| 13 | Breast Cancer Classification with Ensembles | 2017 | Ensemble Learning, Medical AI |
| 14 | TDA in Human Activity Recognition | 2021 | TDA, HAR, Wearables |
| 15 | TDA for Feature Extraction 🏆 | 2021 | TDA, Feature Engineering, Best Paper |
ask-my-ai-clone/
├── README.md
├── requirements.txt
├── .env.example
├── app/
│ ├── main.py # Streamlit chatbot UI
│ ├── config.py # Configuration settings
│ └── components/
│ ├── chat.py # Chat interface
│ ├── sidebar.py # Paper browser sidebar
│ └── citations.py # Citation display component
├── scripts/
│ ├── index_papers.py # PDF → chunks → embeddings
│ ├── build_knowledge.py # Build vector store
│ └── evaluate.py # Test question accuracy
├── papers/ # PDF files of publications
│ ├── 01_nsga_moalo_moda.pdf
│ ├── 02_composite_plates.pdf
│ └── ...
├── data/
│ ├── chroma_db/ # Vector store
│ ├── metadata.json # Paper metadata
│ └── sample_questions.json # Test questions
└── .github/
└── workflows/
└── deploy.yml # Auto-deploy to HF Spaces
Want to create your own AI clone? Fork this repo and:
- Replace papers: Drop your PDFs in
papers/ - Update metadata: Edit
data/metadata.jsonwith your paper details - Re-index: Run
python scripts/index_papers.py - Customize persona: Edit the system prompt in
app/config.py - Deploy: Push to your Hugging Face Space
# app/config.py — Customize the AI persona
SYSTEM_PROMPT = """
You are an AI research assistant trained on the publications of
Prof. Milan Amrut Joshi. You answer questions about:
- Topological Data Analysis & Persistent Homology
- Multi-Objective Optimization (NSGA-II, MOALO, MODA)
- Machine Learning (SVM, PSO, Ensembles)
- Functional Analysis & Pure Mathematics
- Healthcare AI & Big Data
Always cite specific papers when answering. Be accurate,
helpful, and acknowledge when a question is outside your
knowledge base.
"""import requests
response = requests.post(
"https://mlnjsh-ask-my-ai-clone.hf.space/api/ask",
json={"question": "What is persistent homology?"}
)
print(response.json()["answer"])We test the clone on 50+ curated questions across all research areas:
┌─────────────────────────────────────────────┐
│ ACCURACY BY RESEARCH AREA │
├─────────────────────────────────────────────┤
│ TDA & Persistent Homology ████████████ 95%
│ Multi-Objective Optimization ███████████ 92%
│ SVM & Machine Learning ██████████ 88%
│ Pure Mathematics █████████ 85%
│ Healthcare AI ██████████ 90%
│ Engineering Applications █████████ 87%
├─────────────────────────────────────────────┤
│ Overall Accuracy ██████████ 90%
│ Citation Accuracy █████████ 85%
│ Hallucination Rate █ < 5%
└─────────────────────────────────────────────┘
- 🐛 Found an inaccurate answer? Open an issue
- ❓ Suggest a test question by opening a PR to
data/sample_questions.json - ⭐ Star this repo if you find it useful!
@software{joshi2026askclone,
title = {Ask My AI Clone: RAG-Powered Research Q\&A},
author = {Joshi, Milan Amrut},
year = {2026},
url = {https://github.com/mlnjsh/ask-my-ai-clone}
}
 Milan Amrut Joshi Project Author |
 LangChain RAG & LLM application framework |
 LlamaIndex Data framework for LLM applications |