Different challenges.
The same care.

Problem

Agents retyped passport and document data into multiple application forms. A single typo could delay or derail an application.

What we built

Passport and document extraction that prepares visa forms automatically, leaving staff to review instead of retype.

Result

20 to 30 minutes saved per application

Owner-reported project results.

Problem

Staff manually handled bookings, pricing questions, and payment screenshots across WhatsApp and Instagram. Replies slowed outside working hours.

What we built

A connected booking and FAQ system with customer records, distance-based pricing, and payment-proof extraction.

Result

Around 12 hours saved per week in operations

Owner-reported project results.

COMPUTER VISION

Problem

Real-world dice rolls needed to be read fairly and verified before affecting the game.

What we built

A computer vision pipeline that detects, tracks, and reads dice rolls in real time, sending verified results straight to a Unity game server.

Result

Verified dice recognition in under 3 seconds

Owner-reported project results.

AI & AUTOMATION

Problem

Keeping a news channel fresh meant manually finding, summarizing, and posting stories.

What we built

An autonomous pipeline that collects, summarizes, enriches, and posts news to Telegram, with an LLM handling relevance filtering.

Result

Low-maintenance operation with automatic recovery

COMPUTER VISION

Problem

Checking every move in recorded backgammon matches by eye was slow and unreliable.

What we built

A deterministic rule engine that validates backgammon move legality from computer-vision input and exports games to XG format.

Result

Deterministic rule checks for recorded moves

AI & AUTOMATION

Problem

Stable, voice-enabled responses for live audiences across languages.

What we built

A bot that replies to live-stream comments in real time using natural-language generation and multilingual text-to-speech.

Result

Stable message parsing under live load

CUSTOM SOFTWARE

Problem

Staff switched between tabs and systems to look up parts while serving customers.

What we built

A browser extension that surfaces automotive part data directly inside employees' existing workflow.

Result

Around 10 minutes saved per lookup, about 2 hours per day

Owner-reported project results.

AI & AUTOMATION

Problem

A cost-efficient, multilingual RSS-to-Telegram automation.

What we built

A pipeline that turns Portuguese-language news feeds into optimized, cost-efficient Telegram posts via LLM summarization.

Result

Scheduled model usage to control processing cost

AI & AUTOMATION

Problem

Connect an existing SDK-based try-on experience to product-aware conversational intelligence.

What we built

An AI backend and vector search layer connecting a virtual beauty try-on experience with conversational product recommendations.

Result

A product-aware backend for virtual try-on

AI & AUTOMATION

Problem

Help employees find applicable benefits through supportive, policy-grounded conversations.

What we built

A retrieval-based assistant helping employees navigate company wellness benefits and launch relevant platform scenarios.

Result

Policy-grounded answers and benefits navigation

AI & AUTOMATION

Problem

Extract appointment details from calls while making uncertain output visible for review.

What we built

An AI workflow turning appointment call recordings into structured fields, with narrowed context, controlled retries, and manual review.

Result

A staged transcription and extraction workflow