Dataset mapping
Intuitive UI for selecting features and target variables. Automatic type inference (numeric, categorical, datetime), delimiter detection, and tenant-isolated storage so your data never mixes with others.
AutoML Intelligence Engine
From raw data to predictive intelligence
CogniQ is a high-performance Machine Learning platform. Upload datasets, perform feature engineering, and train state-of-the-art models—powered by a distributed Python backend.
Why CogniQ?
Building ML the traditional way means stitching together data pipelines, choosing algorithms, tuning hyperparameters, and managing infrastructure. CogniQ gives you one place to go from raw data to a trained, evaluable model—without sacrificing control or quality.
No pipeline archaeology
Upload your CSV. We handle streaming ingestion, schema inference, and type detection (numeric, categorical, datetime). Your data stays tenant-isolated and ready for the engine.
Algorithm selection, done right
We don’t guess—we run a tournament. XGBoost, LightGBM, CatBoost, and Random Forest compete on your chosen metric (precision or recall). The best model wins; if needed, we tune the top two with Optuna.
Transparent and auditable
Real-time logs, a clear leaderboard, and a summary for every run (SUCCESS, FAILED, or LOW_CONFIDENCE). Test in the playground and monitor accuracy and drift so you know what you’re shipping.
One platform. End-to-end ML.
CogniQ bridges the gap between your data and production-ready models. No PhD required—just your dataset and your goal.
Upload & map
Stream large CSVs, infer schemas automatically, and select features and target in an intuitive UI.
Train at scale
XGBoost, LightGBM, CatBoost, Random Forest—with hyperparameter tuning and metric-driven selection.
Deploy & monitor
Test models in the playground, track accuracy and drift, and manage compute with usage analytics.
How it works
From upload to trained model in a clear, repeatable flow. No black boxes—you see each step and stay in control.
-
Upload your dataset
Drop a CSV (we support large files via streaming). The platform infers delimiters and column types automatically and stores metadata so the engine knows what to use.
-
Map features and target
In the dashboard, select which columns are features and which is the target variable. Optionally exclude columns. The mapping is saved and passed to the training job.
-
Choose your metric
Optimize for precision (e.g. spam, fraud) or recall (e.g. detection, risk). The tournament ranks all algorithms by this metric and picks the winner.
-
Train and watch
Kick off a training job. The Python engine runs preprocessing, then the algorithm tournament (and optional Optuna tuning). You get live log streaming and a final leaderboard.
-
Test and monitor
Use the model playground to run inference with sample inputs. Track accuracy, drift, and compute usage so you can iterate or scale with confidence.
-
Integrate with your systems
Once a model is deployed, integrate CogniQ into your own apps. Use the APIs we provide with your API credentials to request predictions—no need to log in every time. Fetch results and display them in your dashboards, products, or internal tools.
Built for serious ML workflows
Real-time training
Live terminal log streaming from the Python engine. See preprocessing, tournament runs, and results as they happen. Every job produces a clear status: SUCCESS, FAILED, or LOW_CONFIDENCE.
Model playground
Test trained models with live inference inputs. Validate predictions before you ship or integrate. Feature importance and leaderboard data help you explain and trust the model.
Usage analytics
Monitor model accuracy, drift, and compute credit consumption. Usage is tracked by plan (compute hours, datasets, training jobs) so you stay in control of cost and quality.
API & integrations
Integrate CogniQ with your systems. We expose APIs so you can request predictions using your API credentials—no need to log in every time. Get predictions from your deployed models and display them in your own dashboards, apps, or workflows.
Who it’s for
CogniQ fits teams that need predictive models without building and maintaining ML pipelines from scratch.
Data & analytics teams
You have tabular data and a business question. Use CogniQ to go from CSV to a ranked model in one flow, with full visibility into preprocessing and algorithm choice.
Product and growth
Churn, conversion, or risk scores—without a dedicated ML team. Configure target and metric, run training, validate in the playground, then wire predictions into your app via our APIs so your product can show live predictions without opening CogniQ.
Engineers and researchers
You want a fast path to a baseline or a production-ready tabular model. CogniQ handles ingestion, tournament, and tuning; you focus on interpretation and integration.
The engine under the hood
CogniQ’s training runs on a stateless Python AutoML engine, built for reliability and scale. It receives data and config from the orchestrator and always returns a structured result.
Algorithms & tuning
First pass: XGBoost, LightGBM, CatBoost, and Random Forest run on your data; the best by your chosen metric wins. If the best score is below 0.90, we run Optuna hyperparameter tuning on the top two and pick the absolute best. Models below the confidence threshold are still saved but flagged as low-confidence so you can decide.
Precision vs recall
You choose the optimization target. Precision minimizes false positives (e.g. spam, fraud). Recall minimizes false negatives (e.g. detection, medical screening). The tournament ranks every algorithm by this metric—no one-size-fits-all.
Preprocessing & robustness
Automatic feature filtering, target isolation, null handling (median/mode), and categorical encoding. Columns with more than 30% missing get a JSON warning. The engine always writes a summary (and on failure, a traceback) so the orchestrator and you never wait in the dark.
Tri-tier architecture
CogniQ runs on a microservices stack designed for reliability and scale.
The Command Center
A sleek dashboard for dataset management and the model playground. React 19, Vite, Tailwind.
The Orchestrator
Java Spring Boot handles security, persistence, job scheduling, and service coordination.
The Brain
Python AutoML engine: XGBoost, LightGBM, CatBoost, Random Forest, and Optuna tuning. Runs as stateless containers on AWS SageMaker.
Frequently asked questions
- What kind of data does CogniQ support?
- CogniQ is built for tabular data: CSV files. We support streaming uploads (including multi-gigabyte files), automatic delimiter and type detection, and clear feature/target mapping. Time series and unstructured data are not in scope for the current engine.
- Do I need to write code?
- No. You upload your CSV, map features and target in the UI, choose precision or recall, and start training. The dashboard shows logs and results. For integration, you can use API keys and our APIs to get predictions from your own systems.
- Can I get predictions from my own system without logging in?
- Yes. We expose APIs for integration. Once you have a trained and deployed model, use your API credentials to call CogniQ from your application—request predictions and display them in your own dashboards, products, or internal tools. No need to log in every time.
- Where does training run?
- The AutoML engine runs on AWS SageMaker as stateless Docker containers. Your data is sent to the engine for the job and the resulting model artifact is stored securely. The rest of the platform (API gateway, auth, data service, orchestrator, billing) can be deployed in your preferred environment.
- How is billing handled?
- Published plans (Free, Pro at $199/month, Enterprise Quantum at $999/month, and Custom) define limits on compute hours, seats, datasets, training jobs, and predictions. See our Pricing page. Online checkout is processed by Paddle as Merchant of Record. You can view usage and billing history in the app.
- What is your refund policy?
- Cancellations and refunds are described in our Refund Policy. New paid subscribers may request a goodwill refund within 14 days of the first charge, subject to the conditions on that page.
- What if my model score is low?
- If the best tournament score stays below 0.90, we still save the model but mark it as low-confidence. You get a clear status and can decide whether to use it, add more data, or adjust features. Failed runs include a traceback so you can fix data or config and retry.
Transparent pricing
COGNITIVE QUANTUM LLC publishes CogniQ plan prices upfront. Full limits and deliverables are on the pricing page.
Free
$0/ mo
Trial capacity to evaluate upload → train → predict
Pro
$199/ mo
Production workloads for small teams
Enterprise Quantum
$999/ mo
Higher limits and priority support
Ready to turn data into decisions?
Join teams who use CogniQ to ship ML without the complexity.
Launch CogniQ →