diff --git a/Hierarchical/README.md b/Hierarchical/README.md
new file mode 100644
index 00000000..d2f40c91
--- /dev/null
+++ b/Hierarchical/README.md
@@ -0,0 +1,11 @@
+# Hierarchical training of stanene force field
+
+After the OTF training is complete, we use MGP force field in LAMMPS for long time-scale simulations.
+Snapshots are picked up from the LAMMPS trajectory and calculated in DFT for validation and re-training.
+
+1. `add_snap` in mgp-"n": Add snapshots from DFT calculations in mgp-"n-1"
+2. `train_hyps`: Train hyperparameters of GP
+3. `build_mgp`: Use optimized hyps to build MGP
+4. `lmp`: Run LAMMPS with MGP
+5. `uncertainties`: Predict uncertainties 
+6. `dft`: Pick up three snapshots for DFT
diff --git a/OTF/README.md b/OTF/README.md
new file mode 100644
index 00000000..e167a1d9
--- /dev/null
+++ b/OTF/README.md
@@ -0,0 +1,12 @@
+# On-the-fly training of the stanene force field
+
+1. `atom_setup.py` constructs the supercell of stanene
+2. `flare_setup.py` initializes parameters for FLARE force field
+3. `qe_setup.py` sets input parameters of Quantum Espresso (QE). Set to use 288 CPUs in total.
+4. `otf_run.py` sets parameters for OTF training and molecular dynamics. Set to run 100 ps in total.
+5. `postprocess.py` analyzes the OTF trajectory and generate a figure of mean square displacement
+
+To launch the training, use
+```
+python otf_run.py
+```
diff --git a/flare/README.md b/flare/README.md
new file mode 100644
index 00000000..4bf6d00b
--- /dev/null
+++ b/flare/README.md
@@ -0,0 +1,3 @@
+## FLARE source code
+
+This is the FLARE package folked from [flare](https://github.com/mir-group/flare)