Portfolio

Intelligent products, working in the real world.

This is Quaxam's work: AI platforms and systems we have already designed, built and put into production for real organizations. Each one solved a concrete problem and delivers measurable results, day after day. Explore them by capability and let's talk about how they fit your operation.

Portfolio in production

Real products, in production and built end to end

Every one of them is already in production: we designed, built and deployed it — from architecture to model, from data to decision — across healthcare, education, the public sector, retail, logistics, manufacturing, finance, agriculture and embedded hardware. We don't sell promises: each card shows what the product does, the results it already delivers and the technology we built it with.

AI & MLAdvanced analyticsComputer VisionNLP & LanguageEmbedded systemsIntelligent agents
Conceptual diagram of Quaxam's engineering capabilities
Quaxam products

Our products, by capability

Everything you see here is already in production. Filter by capability and open each product: what it does, the results it already delivers and the technology we built it with. If one fits your challenge, we adapt it to your operation.

CapabilitiesAI models compose each protocol (carrier frequencies, binaural difference, phase and duration) optimized against the cortical-entrainment literature, and fine-tune it in real time with reinforcement learning (contextual bandits) and online adaptation: each session learns from the user's response and recalibrates the next.

ResultsStable cortical entrainment from ~4 minutes. Sessions aren't static: the AI tunes them to each person's preferences and outcomes, so the more they're used, the better they hit the mark. A personalization system that improves with every interaction.

In open betaAvailable in the browser and continuously refined: every interaction feeds the AI engine that personalizes the next session. Try it at neuroweave.quaxam.com ↗

Technologies
Contextual banditsReinforcement learningOnline adaptationCortical entrainment
Open NeuroWeave

CapabilitiesIngests vital signs, lab results and notes in real time, computes an hour-by-hour risk score and fires explained alerts (SHAP) inside the clinical workflow.

ResultsAnticipates deterioration hours in advance and delivers every alert prioritized and with the reason why, so the team can act in time.

In productionIn use with a clinical team, integrated into their workflow and their electronic health record.

Technologies
PyTorchTemporal FusionXGBoostSHAPMLflow
Request a demo

CapabilitiesConnected to the LMS and the academic system, it generates weekly risk scores and follow-up dashboards with a factor-by-factor explanation of each case.

ResultsFlags at-risk students weeks in advance and helps prioritize who to support first, with the threshold each institution decides.

In productionIn use at an educational institution, supporting the work of its well-being and retention teams.

Technologies
scikit-learnLightGBMdbtDashboardsCalibration
Request a demo

CapabilitiesReal-time personalized storefront, search and "you might also like", with APIs ready for web, app and email, and built-in A/B experimentation.

ResultsClearly improves recommendation relevance over a popularity-based storefront, and responds in real time even at traffic peaks.

In productionIn use at a retailer, personalizing its online store and its email campaigns.

Technologies
EmbeddingsFAISSTwo-TowerLightFMReal-time APIs
Request a demo

CapabilitiesForecasts per SKU and point of sale with confidence intervals, sensitive to seasonality, price and promotions, with hierarchical reconciliation SKU → category → store.

ResultsSubstantially reduces forecast error, cutting both stockouts and overstock in replenishment.

In productionIn use at a retailer, supporting its purchasing and replenishment cycle per product and point of sale.

Technologies
Time seriesLightGBMHierarchical forecastingOptimizationPipelines
Request a demo

CapabilitiesConsolidates transactions, models treasury and cash-flow scenarios and detects anomalies — all in interactive risk and liquidity dashboards.

ResultsFlags atypical operations and projects cash flow by scenario, with dashboards that update near-live.

In productionIn use at a financial-sector organization, supporting its treasury and risk work.

Technologies
Risk modelsIsolation ForestDuckDBETLVisualization
Request a demo

CapabilitiesReads the document, classifies its type, extracts the key fields (key-value) and validates with rules before exporting to your ERP or management system.

ResultsExtracts fields with strong accuracy and processes in seconds what used to take minutes per document, leaving only doubtful cases for review.

In productionIn use in an organization's back office, moving its documents into the ERP with no manual re-entry.

Technologies
OCRLayoutLMv3NLPLLMsRule-based validation
Request a demo

CapabilitiesQuantized (int8) vision and signal models for Raspberry Pi, Jetson or microcontrollers, with local inference and OTA updates.

ResultsRuns real-time vision models on modest hardware and fits even on a microcontroller, eliminating cloud latency and cost.

In productionInstalled on plant equipment and at sites without stable connectivity, running local inference with OTA model deployment.

Technologies
TinyMLint8 quantizationTensorFlow LiteONNXEdge Computing
Request a demo

CapabilitiesIn-line camera inspection: detects known defects and never-before-seen anomalies, marks the affected area and shows a live quality dashboard.

ResultsCatches both known defects and novel anomalies and leaves piece-by-piece traceability, without slowing the line down.

In productionDeployed on a production line, checking parts at line speed and logging every decision.

Technologies
Computer visionYOLOPatchCoreEdge AIOpenCV
Request a demo

CapabilitiesSearches your documents and systems, answers with source citations and executes actions (function calling) with permissions, traces and guardrails.

ResultsAnswers grounded in the source and always cites where the answer comes from, cutting the time the team wastes searching.

In productionIn internal use at a services organization, connected to its document repositories and management systems, with role-based permissions.

Technologies
LLMsRAGRe-rankingFunction callingVector databases
Request a demo

CapabilitiesIntegrates your sources into a data warehouse, models versioned metrics (dbt) and serves them in dashboards with role-based access control.

ResultsResolves queries fast and refreshes models incrementally, with numbers every department reconciles.

In productionIn use as an organization's analytics layer, consolidating its sources into role-governed dashboards.

Technologies
Data WarehousedbtELTDuckDBDashboards
Request a demo

CapabilitiesMeasures intent and sentiment per candidate on Twitter and Facebook and fuses it with the polling series into a daily forecast, with trend dashboards.

ResultsIn Colombia's 2018 first-round presidential election it achieved an RMSE of 2.47%, below the leading polling firms, and estimated abstention with a 1.72% error.

PublicationPeer-reviewed work: Callejas-Cuervo & Vélez-Guerrero (2019), Revista Politécnica 15(30). Read the paper ↗

Technologies
NLPSentiment analysisTwitter/Facebook APIHybrid modelPython
Discuss a project

CapabilitiesProfiles companies by performance and capabilities, segments them with clustering (k-means / HDBSCAN) and maps the ecosystem's relationships with network analysis.

ResultsReveals the cluster's real segments and the competitiveness gaps between them, turning perception into evidence for prioritizing investment.

ContextModeled on the small-business agro-industrial fabric of Tunja-Boyacá characterized in the regional literature. See the study ↗

Technologies
Clusteringk-means / HDBSCANNetwork analysisGeoanalyticsPython
Discuss a project

CapabilitiesEmbedded hardware designed by us (custom carrier board + SoM module with NPU/Edge-TPU): captures camera and sensor input, runs quantized vision and signal models on-device and delivers results ready to integrate.

ResultsSustains real-time object detection at very low power and keeps inference local, sending no data to the cloud and operating even without a connection.

In productionDeployed in industrial and field equipment, integrating our board, firmware and models into a single device, with OTA model updates.

Technologies
NPU / Edge TPUJetson / SoMCustom PCBTensorRTEmbedded firmware
Request a demo

CapabilitiesLow-power embedded nodes measure soil moisture, temperature, relative humidity and radiation, transmitting over LoRa to a gateway that models the time series and forecasts variable by variable.

ResultsProjects the overnight minimum temperature and soil moisture early enough to foresee frost risk and protect the crop in time.

In productionOperating on high-altitude plots of the Andean plateau; the nodes run on battery and solar panel and report without interruption through the whole season.

Technologies
Embedded sensingLoRa / LoRaWANTime seriesEdge + cloudSolar power
Discuss a project
Measurable impact

What they deliver in operation

Results from real deployments under non-disclosure agreements: the number and the sector, without revealing identities.

−38%

Asset inspection time

Computer vision on the plant floor to catch failures before they escalate.

Energy sector · Colombia

−64%

Document processing time

Intelligent back-office automation and document classification.

Public sector

×3

Decision speed

Analytics and dashboards that replace manual reports with live data.

Retail · Logistics

99.9%

Production uptime

Custom SaaS platforms, monitored and maintained end to end.

In-house platform

Want one of these running in your organization?

They're already proven in production. We adapt any of them to your operation — or build the one you need from scratch. Book a conversation and we'll show you a live demo.