Selected engineering

Anonymized case studies.

Most of our strongest commercial work was developed under confidentiality agreements. We can discuss architecture, engineering decisions, technical challenges and direct contribution — but not proprietary client information.

Case studies

3D Reconstruction Platform

Computer Vision
LiDAR
Photogrammetry
Machine Learning

Problem: Physical environments needed to become measurable digital assets, captured with ordinary devices rather than specialised rigs.
System: A capture-to-reconstruction pipeline turning images and depth data into usable 3D representations.
Engineering approach: Photogrammetry combined with LiDAR depth and learned refinement, with processing split into asynchronous stages so heavy work never blocked the client application.
Technical challenge: Inconsistent capture conditions: lighting, motion, incomplete coverage and noisy depth all degrade reconstruction quality, so the pipeline had to fail gracefully and guide better capture.
Role: End-to-end engineering of the processing pipeline and its integration with the product.
Outcome / use case: Real-world spaces and objects become digital references teams can inspect, measure and reuse.

AI Aesthetic Simulation

Computer Vision
Machine Learning
Facial Analysis

Problem: Professionals needed to show a plausible visual outcome of a localized procedure before it happened.
System: A vision system that identifies facial regions and sub-regions and manipulates them in a controlled, reversible way.
Engineering approach: Region detection and segmentation feeding parametric manipulation, so each change stays bounded to an anatomically meaningful area instead of a generic filter.
Technical challenge: Believability: the simulation must respect identity and anatomy, and small artefacts destroy trust instantly.
Role: Computer vision pipeline and simulation logic.
Outcome / use case: A conversation between professional and client based on something visual rather than verbal description.

Operational Messaging Agent

AI Agents
Tool Calling
Payments
Automation

Problem: Routine operational requests arrived through messaging and consumed hours of manual handling every day.
System: A conversational agent that understands intent and executes real actions inside the company's systems.
Engineering approach: Structured tool calling over a permissioned action layer: scheduling, record updates, payment link generation and human escalation, each with explicit validation and audit trail.
Technical challenge: Reliability under ambiguity — an agent that acts must know when it is not confident enough to act.
Role: Agent architecture, tool layer and integrations.
Outcome / use case: Operational requests are resolved conversationally, with humans handling exceptions instead of routine.

AI Negotiation System

AI Agents
Negotiation
Decision Systems

Problem: Negotiation decisions were made intuitively, without a consistent way to evaluate alternatives and concessions.
System: A specialized AI system grounded in established negotiation frameworks such as BATNA.
Engineering approach: Domain reasoning encoded as explicit structure — alternatives, reservation values, concession paths — with the model reasoning inside that frame rather than free-forming advice.
Technical challenge: Keeping the system grounded: generic language models drift toward agreeable answers, which is the opposite of useful negotiation support.
Role: System design, reasoning structure and evaluation.
Outcome / use case: Negotiation preparation becomes analytical and repeatable instead of improvised.

Document Intelligence System

OCR
Document AI
Multimodal

Problem: Critical business information was locked inside documents, handwriting and specialized records.
System: An extraction and understanding pipeline that turns documents into structured, verifiable data.
Engineering approach: OCR combined with multimodal understanding and schema-constrained extraction, plus confidence scoring so low-certainty fields route to human review.
Technical challenge: Handwriting, scans and non-standard layouts break naive extraction; the system needed to degrade into review rather than into wrong data.
Role: Pipeline architecture, extraction logic and review workflow.
Outcome / use case: Documents stop being an archive and become an operational data source.

Interactive Streaming Platform

Real-time
Streaming
WebSockets
Scale

Problem: Thousands of people needed to participate in the same live moment without perceptible delay.
System: A real-time platform with live interaction, in-stream games and continuously updated rankings.
Engineering approach: Event-driven backend with persistent socket connections, back-pressure handling and state fan-out designed for simultaneous participation rather than simple broadcast.
Technical challenge: Concurrency peaks are instant, not gradual — the system had to hold its behaviour when everyone acts at the same second.
Role: Real-time backend and interaction systems.
Outcome / use case: Live audiences participate instead of only watching.

Retail Intelligence Platform

Geolocation
Crowdsourcing
Mobile
Data

Problem: Price and availability information changes locally and constantly, and no single source keeps up with it.
System: A crowdsourced, location-aware platform that collects, validates and surfaces retail data.
Engineering approach: Mobile capture feeding a geospatial data model, with contribution scoring and validation rules to keep crowdsourced input trustworthy.
Technical challenge: Data quality is the product: without validation, crowdsourced information becomes noise very quickly.
Role: Backend, data model, geolocation and mobile integration.
Outcome / use case: Local commercial reality becomes visible and queryable in near real time.

No client identities, logos, metrics or proprietary details are published. Specifics can be discussed directly, within the limits of each agreement.

Have a hard problem?

Let's figure it out.

Tell us what you're trying to build, what isn't working, or simply where you believe technology could change the outcome.

Start a conversation ceo@galdcode.com

Worldwide • Remote • Open to embedded and international engagements