Impactful cases
Applied AI with measurable results
Each case shows real impact on conversion, cost and performance.
Coven — Marketing AI Control Tower
Multi-tenant SaaS platform for operating marketing with AI agents, human approvals, governance policies, social channels, Meta Ads, Content Velocity Engine, billing, and operational readiness.
Modules implemented
8
auth, billing, operator, social, Meta Ads, content velocity, chat, workspace
Jest tests
91
passing on last verified local run
Mapped flows
6
onboarding → operator → social → billing → approvals → readiness
Challenge
Marketing operations want to leverage AI without losing control: they need to approve actions, track costs, prevent irresponsible automation, connect real channels, and prove results with an auditable trail.
Solution
Coven centralizes operations in a Control Tower with an operator core (runs, actions, policies, approvals), governed Social and Meta Ads capabilities, a Content Velocity Engine, multi-tenant billing, and self-serve onboarding.
Key outcomes
- Multi-tenant SaaS with separate Next.js frontend and FastAPI backend
- Real flows for auth, onboarding, billing, operator, social, Meta Ads, and content velocity mapped and implemented
- Test suite with 91 Jest cases passing and Playwright last run recorded as passed
- Docker build completed successfully in local environment
Smart Engine Sales — Multi-Agent Sales Engine
Sales engine with agents specialized per funnel stage: automatic lead qualification, WhatsApp engagement, LangGraph orchestration, and AI-assisted sales pipeline.
Funnel stages
5
qualification → engagement → follow-up → proposal → closing
Specialized agents
4
qualifier, attendant, follow-up, closer
Integrated channels
2
WhatsApp API + internal interface
Challenge
Sales operations lose leads due to slow response and inconsistent follow-up. Manual qualification is expensive and does not scale — teams waste time on out-of-profile leads.
Solution
Specialized agents per stage (qualification, engagement, follow-up, closing) orchestrated via LangGraph, with human supervision at critical decision points and native WhatsApp integration.
Key outcomes
- Multi-agent pipeline with LangGraph orchestration implemented
- WhatsApp API integration for automated engagement
- ICP qualification with rule-based and LLM scoring
- Configurable human handoff at closing stages
Agency Lead Ops — AI B2B Prospecting Pipeline
B2B prospecting pipeline with data enrichment, ICP scoring, automated outbound sequences, and integrated CRM — acquisition operation with editorial supervision.
Enrichment sources
2
Apollo.io + LinkedIn API
ICP scoring dimensions
4
industry, size, role, buying signals
Pipeline stages
5
discovery → enrich → score → sequence → CRM
Challenge
Manual B2B prospecting is slow, inconsistent, and does not scale. Finding, enriching, and contacting qualified leads takes hours of manual work per prospect.
Solution
AI agents execute enrichment via Apollo.io and LinkedIn, ICP fit scoring, personalized contact sequence creation, and an integrated CRM with human control over message approval.
Key outcomes
- Enrichment pipeline with Apollo.io and LinkedIn API integrated
- ICP scoring across multiple dimensions (industry, size, role, buying signals)
- Outbound sequences with LLM-based personalization
- Internal CRM with per-lead status tracking
MiningOps Copilot — AI for Mining Operations
Operations copilot for mining with local LLM, equipment monitoring, anomaly analysis, and automated operational report generation — without exposing data to the cloud.
Data inference
100% local
sensor data never leaves the perimeter
Supported sensor types
4
vibration, temperature, pressure, flow
Output formats
3
report, alert, query response
Challenge
Mining operations generate massive sensor data volumes, but analysis is slow, manual, and dependent on scarce specialists. Critical operational data cannot leave the site perimeter.
Solution
Local LLM via Ollama processes sensor data in real time, detects anomalies, generates operational reports, and answers natural language queries — on-premise infrastructure with TimescaleDB for time-series data.
Key outcomes
- Local inference with data inside the operational perimeter (no cloud)
- Sensor data anomaly analysis with contextualized LLM
- Automated operational report generation in natural language
- Field operator query interface in Portuguese
Mobility Ad Network — In-Vehicle Advertising Platform
Advertising network on in-vehicle tablets (rideshare model) with geo-targeted ads, campaign management, and audience analytics — turning passenger journeys into media inventory.
Targeting dimensions
3
route, neighborhood, audience profile
External integrations
2
Meta Ads API + Google Maps API
Management surfaces
3
campaign, tablet, analytics
Challenge
In-vehicle media inventory is fragmented and lacks intelligent targeting. Advertisers have no visibility into real audience data, and operators have no platform to manage multiple tablets and campaigns.
Solution
AdTech platform with targeting by route, neighborhood, and audience profile, centralized campaign management for multi-tablet networks, real-time analytics, and Meta Ads API integration for lookalike and retargeting.
Key outcomes
- Campaign management platform for in-vehicle tablet network
- Geo-targeted ads by route, neighborhood, and traffic zone
- Audience analytics with estimated reach per campaign
- Meta Ads API integration for audience synchronization
PRISMA — Clinical AI for Oncology
Oncology-focused clinical reasoning prototype with local LLM via Ollama, semantic literature search (RAG), and auditable case analysis workflow.
Data inference
100% local
privacy model documented
Local latency (doc)
30-50ms
technical comparison vs external APIs
Demo cases
2 cases
liver and lung oncology
Challenge
Deliver AI support for clinical teams without exposing sensitive patient data to external APIs while preserving analysis speed.
Solution
Local-first architecture with Llama 3 via Ollama, ChromaDB for medical literature retrieval, and fallback behavior when AI services are unavailable.
Key outcomes
- Local inference keeps sensitive data inside clinical perimeter
- End-to-end flow: patient data → literature → recommendation
- Two complete oncology demo cases for validation
- Operational fallback when Ollama/ChromaDB are offline
Bellatrix System — Integrated Financial Operations
Flask-based financial operations platform with real-time dashboarding, Nuvemshop integration, shipping/receivables automation, and CI/CD deployment.
Imported sales
140
Nuvemshop history
Migrated installments
222
linked finance records
Imported records
512
validated in deployment docs
Challenge
Unify finance and e-commerce operations in one system while preserving historical data and reliable synchronization of orders, sales, and installments.
Solution
Delivered core financial modules plus Nuvemshop integration with historical import, background sync, and production verification routines.
Key outcomes
- Stable v1.1.0 release with automated deployment
- Nuvemshop historical import completed in production
- Operational dashboard for sales, expenses, and cashflow
- Health-check and environment sync playbooks in place
AutoGrid — Open Source Trading Bot SaaS
Open-source platform for trading bots with Grid and DCA strategies, multi-exchange support, web dashboard, and Telegram execution/alerts.
Supported exchanges
3
Binance, MEXC, Bybit
Native strategies
2
Grid + DCA
License
MIT
fully auditable project
Challenge
Build a trading automation product that is simple for beginners while remaining transparent and customizable for advanced users.
Solution
FastAPI + Next.js stack with bot engine, backtesting (Sharpe/drawdown), CLI tooling, and CCXT integrations for Binance, MEXC, and Bybit.
Key outcomes
- Auditable open-source product under MIT license
- Grid and DCA bot execution through no-code dashboard
- Complete operation surface across API, dashboard, CLI, and Telegram
- Security layer with AES-256-GCM encryption and kill switch
