Extract + Analyse

14 agents in this use case. Each section shows what the agent does, what it needs, its steps in order and an example task.

Causal Analysis#

Estimate the effect of a treatment, programme or policy from observational data. I will attach a data file (CSV, XLSX or SPSS) and say what affects what. Return a report that draws the causal graph and chooses what to control for, picks a design that fits the data (doubly robust adjustment, difference-in-differences, regression discontinuity or an instrument), checks its assumptions, estimates the effect with a confidence interval and shows how strong unmeasured confounding would need to be to explain it away, plus the estimates and balance in a spreadsheet.

Steps

START → [1] Causal question and graph → [2] Strategy and analysis data →
  [3] Assumption checks and estimate → [4] Sensitivity and report → [5] Answer → END

Descriptive Statistics#

Summarise my data for a manuscript. I will attach a data file (CSV, XLSX or SPSS), optionally with a codebook, and say which variables or groups matter. Return a report with a Table 1 of sample characteristics, summaries chosen for each variable's type (mean and SD or median and quartiles, counts and percentages), group columns with standardised mean differences, missing data, distribution figures and a sample-description paragraph, plus the tables in a spreadsheet.

Steps

START → [1] Data check and summary plan → [2] Cleaning and summaries →
  [3] Group balance and figures → [4] Sample description and report → [5] Answer → END

ECG HRV Analysis#

Analyse heart rate variability from my ECG recordings. I will attach ECG files (CSV, EDF, BDF and others), optionally with an events or participants table. Return a report with signal quality, corrected R peaks, time, frequency and nonlinear HRV per condition, condition and group comparisons with effect sizes and figures and a metrics table. Each step shows the code it ran.

Steps

START → [1] Data check and signal quality → [2] Preprocessing and HRV metrics →
  [3] Comparisons and figures → [4] Interpretation and report → [5] Answer → END

EDA Analysis#

Analyse electrodermal activity (skin conductance) in my recordings. I will attach EDA files (CSV, Empatica E4 EDA.csv, EDF), optionally with stimulus onsets or condition blocks in an events table and a participants table. Return a report with signal quality, tonic and phasic components, skin conductance level, non-specific and event-related skin conductance responses per condition, condition and group comparisons with effect sizes and figures, and a metrics table. Each step shows the code it ran.

Steps

START → [1] Data check and signal quality → [2] Preprocessing and EDA metrics →
  [3] Comparisons and figures → [4] Interpretation and report → [5] Answer → END

EEG ERP Analysis#

Measure event-related potentials in my EEG. I will attach EEG recordings with stimulus markers (EDF, BDF, FIF, BrainVision, EEGLAB or CSV), optionally with an events table and what the event codes mean. Return a report with signal quality, preprocessing (filtering, interpolation, reference, ICA, epochs and baseline, trial rejection), component mean and peak amplitude and latency per condition, condition and group comparisons with effect sizes, grand-average waveforms, difference waves and topographies, and a metrics table. Each step shows the code it ran.

Steps

START → [1] Data check and signal quality → [2] Preprocessing, epochs and components →
  [3] Comparisons and figures → [4] Interpretation and report → [5] Answer → END

EEG Spectral Analysis#

Analyse resting or task EEG in the frequency domain. I will attach EEG recordings (EDF, BDF, FIF, BrainVision, EEGLAB or CSV), optionally with an events or participants table. Return a report with signal quality and bad channels, preprocessing (filtering, interpolation, average reference, ICA), band power, peak alpha frequency, frontal alpha asymmetry and the 1/f exponent per condition, condition and group comparisons with effect sizes, topographies and spectra, and a metrics table. Each step shows the code it ran.

Steps

START → [1] Data check and signal quality → [2] Preprocessing and spectral metrics →
  [3] Comparisons and figures → [4] Interpretation and report → [5] Answer → END

Exploratory Data Analysis#

Explore my data before I decide what to analyse. I will attach a data file (CSV, XLSX or SPSS), optionally with a codebook and the variables I care about. Return a report that ranks data-quality problems, describes missing-data patterns, distributions and outliers, relationships between variables (including ones that change direction within subgroups) and hidden groupings of cases, and turns the findings into hypothesis candidates with the next analysis for each, plus a data profile spreadsheet.

Steps

START → [1] Data check and quality → [2] Missing data and distributions →
  [3] Relationships and structure → [4] Hypothesis candidates and report → [5] Answer → END

Facial Expression Analysis#

Analyse facial expressions in face videos or photos — one person or many participants. I will attach the files (MP4, MOV, WEBM, JPG, PNG; several as a ZIP), optionally with segment timings in an events table and a participants table. For one person, return the overall expression distribution (dominant expressions, emotion scores, valence and arousal, action-unit profile) and the change over time (time bins, trends, expression episodes, peak moments) with segment comparisons; for several participants, compare conditions and groups relative to each participant's baseline with effect sizes. Figures contain no face images; a metrics table comes with the report. Each step shows the code it ran.

Steps

START → [1] Data check and quality → [2] Expression extraction and measures →
  [3] Analysis and figures → [4] Interpretation and report → [5] Answer → END

fMRI Analysis#

Analyse my fMRI data from raw BOLD runs. I will attach a BIDS dataset or 4-D NIfTI runs with event files (or resting-state runs), optionally with participants.tsv. Return a report with image quality and head motion, motion correction and MNI registration, task activation (general linear model) or resting-state connectivity, whole-brain group maps with FDR-corrected clusters, network-level comparisons with effect sizes and figures, and a metrics table. Each step shows the code it ran.

Steps

START → [1] Data check and image quality → [2] Preprocessing and first level →
  [3] Group analysis and figures → [4] Interpretation and report → [5] Answer → END

Inferential Statistics#

Test my hypotheses on my data. I will attach a data file (CSV, XLSX or SPSS) and state my hypotheses or the comparison I want, optionally with a codebook or analysis plan. Return a statistical report that, for each hypothesis, explains why the test was chosen, the assumption checks, the result with effect size, 95% CI and an APA sentence, corrected for multiple comparisons, plus a results table.

Steps

START → [1] Data check and analysis plan → [2] Cleaning and assumption checks →
  [3] Tests, effect sizes and robustness → [4] Interpretation and report → [5] Answer → END

PPG Pulse Analysis#

Analyse pulse rate and pulse-rate variability from my PPG recordings, including wearables. I will attach PPG or BVP files (CSV, Empatica E4 BVP.csv with ACC.csv, EDF), optionally with accelerometer columns and an events or participants table. Return a report with pulse quality and motion, corrected pulse peaks, time, frequency and nonlinear pulse-rate variability per condition, condition and group comparisons with effect sizes and figures, and a metrics table. Each step shows the code it ran.

Steps

START → [1] Data check and signal quality → [2] Preprocessing and pulse-rate variability →
  [3] Comparisons and figures → [4] Interpretation and report → [5] Answer → END

Predictive Modeling#

Build and compare prediction models for an outcome in tabular data. I will attach a data file (CSV, XLSX or SPSS) and say what to predict. Return a report that checks for leakage, compares a baseline, penalised regression, random forest and gradient boosting on the same cross-validation splits with confidence intervals, checks calibration and picks a threshold, shows which predictors drive the predictions and how performance differs by subgroup, with a TRIPOD+AI checklist, plus the performance and out-of-fold predictions in a spreadsheet.

Steps

START → [1] Prediction problem → [2] Leakage check and baseline →
  [3] Model comparison and calibration → [4] Contribution and report → [5] Answer → END

Prescriptive Analysis#

Turn analysis results into a decision. I will attach an earlier analysis result or a data file and state the options, the goal and the constraints (costs, budget, capacity). Return a report that grades the evidence behind each option (experiment, causal estimate, prediction or assumption), compares the options by expected net benefit or finds the best allocation of a limited resource, simulates the uncertainty to show how likely the recommendation is best and what it risks, and shows which inputs would flip the decision, plus the scenarios in a spreadsheet.

Steps

START → [1] Decision problem → [2] Evidence and value model → [3] Decision and uncertainty →
  [4] Sensitivity and report → [5] Answer → END

Remote PPG (Face Video Heart Rate)#

Estimate heart rate from my participants' face videos (remote PPG). I will attach face videos, optionally with a contact heart-rate file per video, an events table and a participants table. Return a report with video quality, heart rate per condition from POS (with CHROM and a deep model for comparison), signal-to-noise ratio and reliable windows, error against the reference when given, condition and group comparisons with effect sizes and figures (no face images), and a metrics table. Each step shows the code it ran.

Steps

START → [1] Video check and quality → [2] Pulse extraction and heart rate →
  [3] Comparisons and figures → [4] Interpretation and report → [5] Answer → END