Raman Suite

Three-tab Raman analysis platform: (1) spectrum matching against JARVIS ramandb (~5K materials), (2) RamanGPT AI-powered structure prediction from Raman peaks, and (3) ALIGNN-based Raman spectrum prediction from crystal structures.

Open App

---

Overview

The Raman Suite provides a complete workflow for Raman spectroscopy in materials science. Match experimental spectra to known materials via cosine similarity, predict crystal structures from Raman peaks using RamanGPT, or predict Raman spectra from POSCAR structures using ALIGNN (jv_raman_alignn). The ramandb contains ~5K materials with DFT-computed Raman tensors, cross-referenced with JARVIS-DFT JVASP IDs.

Data Source

ramandb (~5K entries with DFT Raman tensors). RamanGPT model (knc6/ramangpt_dft_model) for AI structure prediction. ALIGNN (jv_raman_alignn) for ML Raman spectrum prediction.

Endpoints

POST /raman/match — Match spectrum against ramandb

Gaussian-broadens user spectrum and computes cosine similarity against all entries in ramandb. Optional formula filter narrows candidates.

curl -X POST "https://atomgpt.org/raman/match" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "frequencies": [135.2, 196.7, 221.3, 257.0, 418.3, 594.7, 650.1, 701.6, 705.6],
    "activities": [0.024, 0.006, 0.013, 0.091, 0.083, 0.340, 0.001, 1.000, 0.137],
    "formula": "SrTeO3",
    "sigma": 8.0,
    "top_n": 10,
    "freq_min": 0,
    "freq_max": 1200
  }'

Match without formula filter (search entire database):

curl -X POST "https://atomgpt.org/raman/match" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "frequencies": [520, 302, 960],
    "activities": [1.0, 0.02, 0.06],
    "formula": "",
    "sigma": 8.0,
    "top_n": 5
  }'

Field	Type	Default	Description
frequencies	list[float]	required	Raman shift values (cm⁻¹)
activities	list[float]	required	Raman activity/intensity values
formula	string	""	Optional formula or element filter (e.g. "Si", "SrTeO3", "Te,O")
sigma	float	8.0	Gaussian broadening width (cm⁻¹)
top_n	int	10	Number of top matches to return
freq_min	float	0	Minimum frequency for matching grid
freq_max	float	1200	Maximum frequency for matching grid

Response includes grid (500-point x-axis), user_spectrum (broadened), and matches array with id, formula, elements, similarity, n_active_modes, active_freqs, active_activities, and spectrum (broadened reference).

---

POST /raman/lookup — Look up a material's Raman spectrum by ID

curl -X POST "https://atomgpt.org/raman/lookup" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "material_id": "JVASP-1002"
  }'

Returns freq_cm, raman_activity, formula, and elements for the given material.

---

POST /raman/ramangpt — AI structure prediction from Raman peaks

Given a chemical formula and Raman peak data, RamanGPT generates a crystal structure (POSCAR) that would produce the observed spectrum.

curl -X POST "https://atomgpt.org/raman/ramangpt" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "formula": "SrTeO3",
    "raman_data": "135.2  0.024\n196.7  0.006\n221.3  0.013\n257.0  0.091\n418.3  0.083\n594.7  0.340\n701.6  1.000",
    "max_new_tokens": 512
  }'

Field	Type	Default	Description
formula	string	required	Chemical formula
raman_data	string	required	Frequency/intensity pairs, newline-separated
max_new_tokens	int	512	Max generation tokens

Response includes poscar, structure (formula, spacegroup, num_atoms), raw_output, prompt, peak_text, inference_time.

---

GET /ramangpt/query — Direct RamanGPT prediction (GET)

curl "https://atomgpt.org/ramangpt/query?formula=SrTeO3&peaks=135.2cm-1(0.024),257.0cm-1(0.091),701.6cm-1(1.000)&max_new_tokens=512" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "accept: application/json"

Param	Description
formula	Chemical formula
peaks	Peaks as freq_cm-1(intensity) comma-separated
max_new_tokens	Max generation tokens (default 512)

Returns plain-text POSCAR of the predicted structure.

---

POST /raman/alignn — Predict Raman spectrum from POSCAR

Uses the ALIGNN jv_raman_alignn model to predict a 200-point Raman spectrum (50–1000 cm⁻¹) from a crystal structure.

curl -X POST "https://atomgpt.org/raman/alignn" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "poscar": "Si2\n1.0\n3.364 0.0 1.942\n1.121 3.172 1.942\n0.0 0.0 3.885\nSi\n2\nCartesian\n3.925 2.775 6.798\n0.561 0.396 0.971"
  }'

Response includes x_axis (200 points, 50–1000 cm⁻¹), y_axis, y_normalized, peaks (top 20 with freq/intensity/normalized), formula, spacegroup, lattice, inference_time.

---

Python Examples

Match spectrumRamanGPTALIGNN RamanLookup by ID

import requests

response = requests.post(
    "https://atomgpt.org/raman/match",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={
        "frequencies": [135.2, 257.0, 418.3, 594.7, 701.6],
        "activities": [0.024, 0.091, 0.083, 0.340, 1.000],
        "formula": "SrTeO3",
        "sigma": 8.0,
        "top_n": 5,
    },
)
data = response.json()
for m in data["matches"]:
    print(f"{m['id']:12s} {m['formula']:10s} sim={m['similarity']:.3f}")

import requests

response = requests.post(
    "https://atomgpt.org/raman/ramangpt",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={
        "formula": "SrTeO3",
        "raman_data": "135.2  0.024\n257.0  0.091\n701.6  1.000",
        "max_new_tokens": 512,
    },
)
data = response.json()
if data["success"]:
    print(f"Formula: {data['structure']['formula']}")
    print(f"Spacegroup: {data['structure']['spacegroup']}")
    print(f"POSCAR:\n{data['poscar'][:300]}")

import requests

SI_POSCAR = """Si2
1.0
3.364 0.0 1.942
1.121 3.172 1.942
0.0 0.0 3.885
Si
2
Cartesian
3.925 2.775 6.798
0.561 0.396 0.971"""

response = requests.post(
    "https://atomgpt.org/raman/alignn",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={"poscar": SI_POSCAR},
)
data = response.json()
if data["success"]:
    print(f"Formula: {data['formula']}, Peaks: {data['n_peaks']}")
    for p in data["peaks"][:5]:
        print(f"  {p['freq']} cm-1  intensity={p['intensity']:.4e}")

import requests

response = requests.post(
    "https://atomgpt.org/raman/lookup",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={"material_id": "JVASP-1002"},
)
data = response.json()
if data["success"]:
    print(f"Formula: {data['formula']}")
    for f, a in zip(data["freq_cm"], data["raman_activity"]):
        print(f"  {f:.1f} cm-1  activity={a:.4f}")

AGAPI Agent [WIP]

from agapi.agents import AGAPIAgent
import os

agent = AGAPIAgent(api_key=os.environ.get("AGAPI_KEY"))

# Match a spectrum
response = agent.query_sync("Match Raman spectrum for SrTeO3: 135 0.02, 257 0.09, 702 1.0")
print(response)

# Predict structure from Raman
response = agent.query_sync("Use RamanGPT to predict the structure of TiO2 from Raman peaks at 144, 399, 639 cm-1")
print(response)

# Predict Raman from structure
response = agent.query_sync("Predict the Raman spectrum of JVASP-1002 using ALIGNN")
print(response)

References

• Computational Raman Database DOI
• npj Comp. Mat. 7, 185 (2021) — ALIGNN DOI
• atomgptlab/atomgpt