diff --git a/codex-plonky2-circuits/src/circuits/mod.rs b/codex-plonky2-circuits/src/circuits/mod.rs
index f4e17f6..744d84d 100644
--- a/codex-plonky2-circuits/src/circuits/mod.rs
+++ b/codex-plonky2-circuits/src/circuits/mod.rs
@@ -1,3 +1,4 @@
 pub mod capped_tree_circuit;
 pub mod safe_tree_circuit;
-pub mod prove_single_cell;
\ No newline at end of file
+pub mod prove_single_cell;
+pub mod sample_cells;
\ No newline at end of file
diff --git a/codex-plonky2-circuits/src/circuits/sample_cells.rs b/codex-plonky2-circuits/src/circuits/sample_cells.rs
new file mode 100644
index 0000000..4b67349
--- /dev/null
+++ b/codex-plonky2-circuits/src/circuits/sample_cells.rs
@@ -0,0 +1,187 @@
+// Sample cells
+// consistent with:
+// https://github.com/codex-storage/codex-storage-proofs-circuits/blob/master/circuit/codex/sample_cells.circom
+// circuit consists of:
+// - reconstruct the dataset merkle root using the slot root as leaf
+// - samples multiple cells by calling the sample_cells
+
+use anyhow::Result;
+use plonky2::field::extension::Extendable;
+use plonky2::hash::hash_types::{HashOut, RichField};
+use plonky2::iop::target::{BoolTarget, Target};
+use plonky2::iop::witness::{PartialWitness, WitnessWrite, Witness};
+use plonky2::plonk::circuit_builder::CircuitBuilder;
+use plonky2::plonk::circuit_data::CircuitConfig;
+use plonky2::plonk::config::{AlgebraicHasher, GenericConfig, Hasher, GenericHashOut};
+use std::marker::PhantomData;
+use itertools::Itertools;
+
+use crate::merkle_tree::merkle_safe::MerkleTree;
+use plonky2::hash::poseidon::PoseidonHash;
+
+use plonky2::hash::hash_types::{HashOutTarget, NUM_HASH_OUT_ELTS};
+use crate::merkle_tree::merkle_safe::{MerkleProof, MerkleProofTarget};
+use plonky2_poseidon2::poseidon2_hash::poseidon2::Poseidon2;
+
+use plonky2::field::goldilocks_field::GoldilocksField;
+use plonky2::plonk::config::PoseidonGoldilocksConfig;
+
+use plonky2::hash::hashing::PlonkyPermutation;
+use crate::circuits::prove_single_cell::{MAX_DEPTH, SlotTree};
+use crate::circuits::safe_tree_circuit::{MerkleTreeCircuit, MerkleTreeTargets};
+
+// constatnts and types
+const DATASET_DEPTH: usize = 8;
+const N_SAMPLES: usize = 5;
+
+type HF = PoseidonHash;
+
+// ------ Dataset Tree --------
+///dataset tree containing all slot trees
+#[derive(Clone)]
+pub struct DatasetTree<F: RichField, H: Hasher<F>> {
+    pub tree: MerkleTree<F,H>, // dataset tree
+    pub slot_trees: Vec<SlotTree<F,H>>, // vec of slot trees
+}
+
+/// Dataset Merkle proof struct, containing the dataset proof and N_SAMPLES proofs.
+#[derive(Clone)]
+pub struct DatasetMerkleProof<F: RichField, H: Hasher<F>> {
+    pub slot_index: usize,
+    pub entropy: usize,
+    pub dataset_proof: MerkleProof<F,H>,       // proof for dataset level tree
+    pub slot_proofs: Vec<MerkleProof<F,H>>, // proofs for sampled slot, contains N_SAMPLES proofs
+}
+
+impl<F: RichField, H: Hasher<F>> Default for DatasetTree<F,H> {
+    /// dataset tree with fake data, for testing only
+    fn default() -> Self {
+        let mut slot_trees = vec![];
+        let n_slots = 1<<DATASET_DEPTH;
+        for i in 0..n_slots {
+            slot_trees.push(SlotTree::<F,H>::default());
+        }
+        // get the roots or slot trees
+        let slot_roots = slot_trees.iter()
+            .map(|t| {
+                t.tree.root().unwrap()
+            })
+            .collect::<Vec<_>>();
+        // zero hash
+        let zero = HashOut {
+            elements: [F::ZERO; 4],
+        };
+        let dataset_tree = MerkleTree::<F, H>::new(&slot_roots, zero).unwrap();
+        Self{
+            tree: dataset_tree,
+            slot_trees,
+        }
+    }
+}
+
+
+impl<F: RichField, H: Hasher<F>> DatasetTree<F, H> {
+    /// same as default but with supplied slot trees
+    pub fn new(slot_trees: Vec<SlotTree<F,H>>) -> Self{
+        // get the roots or slot trees
+        let slot_roots = slot_trees.iter()
+            .map(|t| {
+                t.tree.root().unwrap()
+            })
+            .collect::<Vec<_>>();
+        // zero hash
+        let zero = HashOut {
+            elements: [F::ZERO; 4],
+        };
+        let dataset_tree = MerkleTree::<F, H>::new(&slot_roots, zero).unwrap();
+        Self{
+            tree: dataset_tree,
+            slot_trees,
+        }
+    }
+
+    /// generates a dataset level proof for given slot index
+    /// just a regular merkle tree proof
+    pub fn get_proof(&self, index: usize) -> MerkleProof<F, H> {
+        let dataset_proof = self.tree.get_proof(index).unwrap();
+        dataset_proof
+    }
+
+    /// generates a proof for given slot index
+    /// also takes entropy so it can use it sample the slot
+    pub fn sample_slot(&self, index: usize, entropy: usize) -> DatasetMerkleProof<F, H> {
+        let dataset_proof = self.get_proof(index);
+        let slot = &self.slot_trees[index];
+        let slot_root = slot.tree.root().unwrap();
+        let mut slot_proofs = vec![];
+        // get the index for cell from H(slot_root|counter|entropy)
+        for i in 0..N_SAMPLES {
+            let cell_index_bits = calculate_cell_index_bits(entropy, slot_root, i);
+            let cell_index = bits_le_padded_to_usize(&cell_index_bits);
+            slot_proofs.push(slot.get_proof(cell_index));
+        }
+
+        DatasetMerkleProof{
+            slot_index: index,
+            entropy,
+            dataset_proof,
+            slot_proofs,
+        }
+    }
+
+    // verify the sampling - non-circuit version
+    pub fn verify_sampling(&self, proof: DatasetMerkleProof<F,H>) -> Result<bool>{
+        let slot = &self.slot_trees[proof.slot_index];
+        let slot_root = slot.tree.root().unwrap();
+        // check dataset level proof
+        let d_res = proof.dataset_proof.verify(slot_root,self.tree.root().unwrap());
+        if(d_res.unwrap() == false){
+            return Ok(false);
+        }
+        // sanity check
+        assert_eq!(N_SAMPLES, proof.slot_proofs.len());
+        // get the index for cell from H(slot_root|counter|entropy)
+        for i in 0..N_SAMPLES {
+            let cell_index_bits = calculate_cell_index_bits(proof.entropy, slot_root, i);
+            let cell_index = bits_le_padded_to_usize(&cell_index_bits);
+            //check the cell_index is the same as one in the proof
+            assert_eq!(cell_index, proof.slot_proofs[i].index);
+            let s_res = slot.verify_cell_proof(proof.slot_proofs[i].clone(),slot_root);
+            if(s_res.unwrap() == false){
+                return Ok(false);
+            }
+        }
+        Ok(true)
+    }
+}
+
+// --------- helper functions --------------
+fn calculate_cell_index_bits<F: RichField>(p0: usize, p1: HashOut<F>, p2: usize) -> Vec<bool> {
+    let p0_field = F::from_canonical_u64(p0 as u64);
+    let p2_field = F::from_canonical_u64(p2 as u64);
+    let mut inputs = Vec::new();
+    inputs.extend_from_slice(&p1.elements);
+    inputs.push(p0_field);
+    inputs.push(p2_field);
+    let p_hash = HF::hash_no_pad(&inputs);
+    let p_bytes = p_hash.to_bytes();
+
+    let p_bits = take_n_bits_from_bytes(&p_bytes, MAX_DEPTH);
+    p_bits
+}
+fn take_n_bits_from_bytes(bytes: &[u8], n: usize) -> Vec<bool> {
+    bytes.iter()
+        .flat_map(|byte| (0..8u8).map(move |i| (byte >> i) & 1 == 1))
+        .take(n)
+        .collect()
+}
+/// Converts a vector of bits (LSB first) into an index (usize).
+fn bits_le_padded_to_usize(bits: &[bool]) -> usize {
+    bits.iter().enumerate().fold(0usize, |acc, (i, &bit)| {
+        if bit {
+            acc | (1 << i)
+        } else {
+            acc
+        }
+    })
+}