Selena Avcı

AI Hub Portal

Central landing portal that aggregates all agents and provides unified navigation, dataset routing, and discovery.

Open Portal →

Overview

AI Hub reduces dependency on data science teams by letting business users upload their own datasets and run analytics, modeling, and content-generation workflows through a collection of modular agents. Every agent is self-contained, transparent in its logic, and exports business-ready outputs (Excel, CSV, JSON, dashboards). The platform combines classical ML (scikit-learn) at its core with optional LLM augmentation where natural-language interpretation adds value.

A/B Testing Agent

Experimentation
  • Three-step experiment workflow (define → upload → decide) with a single decision card output: ship / hold / continue / no-effect / redesign.
  • Deterministic statistical engine: frequentist tests, confidence intervals, guardrails, segment analysis, and Bayesian probability.
Hypothesis Testing Confidence Intervals Bayesian Inference Guardrail Metrics Segment Analysis Streamlit
Open on Streamlit Cloud →

Anomaly Detection Agent

Detection
  • Runs three unsupervised models in parallel and auto-selects the best by silhouette score; produces per-record anomaly scores and feature-level deviation explanations.
  • Highlights flagged rows in an exportable Excel report ready for risk, audit, and operational review.
Isolation Forest One-Class SVM Local Outlier Factor Silhouette Score Streamlit
Open on Streamlit Cloud →

Data Cleaning Agent

Preparation
  • Detects missing values, duplicates, type mismatches, placeholder values, format issues, and outliers; user selects the corrective action per issue.
  • Applies user-approved transformations sequentially and emits a transparent change log so the cleaned dataset feeds downstream agents safely.
pandas numpy Missing Value Imputation IQR / Winsorize Type Coercion Audit Log
Open on Streamlit Cloud →

Data Comparison Agent

Reconciliation
  • Compares 2–4 files (CSV / Excel / XML) on a key column; surfaces row-level diffs with yellow / green / red color coding and presence-matrix across files.
  • Metric mode computes percent change, absolute delta, and ratio against a chosen baseline; read-only by design, full Excel export with preserved color encoding.
Key-based Matching Diff Engine Percent Change Read-only Pipeline Excel Export Optional LLM Narrative
Open on Streamlit Cloud →

Data Quality Agent

Quality Control
  • Detects 8 issue types (missing, format, type drift, semantic inconsistency, duplication, range violation, sparse columns, low-signal features) and assigns Low → Critical risk levels.
  • Cell-level color highlighting and prioritized issue list ensures downstream agents (anomaly, segmentation, forecasting) operate on validated data.
Rule-based Validation Risk Scoring Statistical Profiling Cell-level Highlighting Excel Reports
Open on Streamlit Cloud →

Data Visualization Agent

Visualization
  • Auto-profiles columns, recommends chart candidates, and produces interactive Plotly visuals (distribution, trend, correlation, comparison).
  • Manual chart-builder flow for users who want full custom control; bulk download of all generated visuals.
Plotly Auto Data Profiling Chart Recommendation Streamlit
Open on Streamlit Cloud →

Feature Engineering Agent

Preparation
  • Auto-detects column types and offers type-appropriate transformations alongside “What is it? / When to use it?” explanations for each option.
  • Every applied transformation is recorded in a timestamped log; output dataset and log are downloadable as CSV / Excel.
Date Decomposition One-Hot / Frequency Encoding Z-score Log Transform Binning Missing-Value Flags
Open on Streamlit Cloud →

Forecasting Agent

Time Series
  • Profiles the series, auto-detects frequency, runs four forecasting models in parallel, and selects the lowest-error model with full transparency on relative performance.
  • What-if scenario overlays for campaigns, special days, and trend assumptions; designed as a decision-support tool, not an autonomous decision engine.
SARIMA Prophet XGBoost Naive Baseline What-if Simulation Confidence Intervals
Open on Streamlit Cloud →

MailCraft Agent

LLM
  • LLM-powered email generation for new emails or replies; tone, length, category, language, salutation, and corporate signature parameters are enforced via system-prompt engineering.
  • One-click variant regeneration (shorter / more formal / more polite / clearer) and bilingual TR/EN support.
LLM Prompt Engineering Tone Control Bilingual TR/EN vLLM Streamlit
Open on Streamlit Cloud →

Mock Data Generator Agent

Synthetic Data
  • Two modes: Upload (auto-extracts schema from a sample file and replicates structure) and Scratch (column-by-column definition from zero).
  • Supports schema consistency, categorical weighting, correlation rules, business rules, and noise modes — output up to 1M rows in CSV / Excel / JSON.
Faker Schema Inference Statistical Sampling Correlation Rules Business Rule Engine CSV / Excel / JSON Export
Open on Streamlit Cloud →

Prompt Optimization Agent

LLM Tooling
  • Refines raw user prompts into clearer, more structured, result-focused instructions tailored to a target agent (Code Assist / MailCraft / General Purpose).
  • Acts as a quality multiplier across the rest of AI Hub, indirectly raising the precision of every LLM-backed agent.
LLM Prompt Engineering Template Library Agent-Aware Routing
Open on Streamlit Cloud →

Quiz Generator Agent

Content Generation
  • Generates context-grounded MCQ, True/False, and Fill-in-the-Blank questions from training documents (PDF, PPTX, DOCX, TXT) using an on-prem / corporate LLM.
  • Document parsing, paragraph-aware chunking, multi-stage LLM generation, quality filters, Turkish-first NFC normalization, and LMS-compatible (Enocta) Excel export.
LLM Document Parsing Smart Chunking Multi-Stage Pipeline NFC Normalization vLLM LMS Excel Export
Open on Streamlit Cloud →

Segment Intelligence Agent

Clustering
  • Performs unsupervised segmentation and surfaces, for each cluster, summary statistics, distinguishing features, and deltas vs. the overall population.
  • Cluster summaries can be translated into business-language profiles and action ideas with LLM.
K-Means Cluster Profiling Feature Differentiation LLM Streamlit
Open on Streamlit Cloud →

Smart Modeling Agent

AutoML
  • Five-tab guided flow: upload → profile → problem definition → train → results. Auto preprocessing (imputation, encoding, optional scaling), stratified split where needed, multi-algorithm comparison.
  • Explicit positioning as a prototyping / exploration tool, not production deployment — every choice is logged into a downloadable Modeling Artifact ZIP for auditability.
scikit-learn Classification Regression Auto Preprocessing Stratified Split Multi-Algorithm Comparison Modeling Artifact
Open on Streamlit Cloud →

Social Pulse AI

Social Intelligence
  • Multi-bank social-media analytics platform: sentiment distribution, engagement, reach, anomaly detection, and topic breakdowns from multi-sheet Excel datasets.
  • Modules for brand perception, crisis early-warning, competitor benchmarking, audience profiling, and action recommendations — turns raw comments into “Insight → Decision → Action”.
LLM Sentiment Analysis Anomaly Detection Competitor Benchmarking Topic Analytics Streamlit
Open on Streamlit Cloud →