All projects

The full picture.

01

FGS Global

Communications · Enterprise AI · Agent evaluation and monitoring

Test · Protect · Monitor
30+
Specialized AI agents
12
Quality scorers
300+
Test cases in the release suite

One scorecard for 30+ agents across different use cases.

FGS Global runs specialist AI agents for M&A, IPO communications, research and stakeholder analysis. We built the evaluation and monitoring layer so every agent is graded the same way and every release is tested before it ships.

What this enabled

  • Productivity, quality and cost graded on one unified scale across use cases
  • Results split by user context, including location and seniority
  • Releases gated on critical checks such as hallucination, relevance and faithfulness
  • Production conversations reviewed and turned into new test cases

02

Sourcera

Procurement · Supplier intelligence · Opportunity evaluation

Test · Prove
Millions
In savings identified
Benchmarked
Against reference datasets
Automated
No manual scoring

Saving opportunities scored against a benchmark, not by opinion.

Sourcera generates cost-saving opportunities from spend and supplier data. We built the evaluation that decides which opportunities are worth acting on. The process now runs without manual effort.

What this enabled

  • Savings identified measured on the opportunities the system generates
  • Output quality compared against benchmark datasets
  • Each opportunity scored on geographic relevance, critical suppliers and cross-domain suppliers
  • Fully automated, so scoring is repeatable across every run

03

Strict

Manufacturing · AI-First ERP Transformation

AI Workflow Systems
2h+
Freed per person daily
15+ Years
Historical data migrated
AI-first
On ERP backbone

An AI-first operational transformation that frees more than 2 hours per person per day

Strict's backoffice handles email, invoicing, and operations across disconnected tools. We are building an AI-first transformation on top of an ERPNext backbone: an AI email assistant that auto-drafts shared inbox replies, an OCR and LLM invoice extraction tool that classifies and counts line items automatically, and a centralised ERP replacing Lexware, Google Sheets, and legacy systems. The combined transformation is projected to free more than 2 hours per person per day and reduce the mental load of context-switching across fragmented tools.

What this enabled

  • AI email assistant drafts shared inbox replies automatically. Staff review and send, the system never does
  • OCR and LLM invoice tool extracts and classifies line items (prints, textiles, non-relevant) into a GDPR-compliant cloud database
  • ERPNext centralises Lexware, Google Sheets, and legacy tools into one workplace, migrating 15+ years of historical data
  • Projected to free 2+ hours per person daily, representing measurable recoverable capacity across the entire backoffice

04

Expresssteuer

Tax Services · Data & AI Engineering

AI Workflow Systems
>90%
Database cost reduction
~70%
Operational cost savings
2
AI solutions deployed

Rebuilding a cloud data foundation and cutting costs with AI-driven automation

Expresssteuer, a tax services company, needed to modernize its cloud data infrastructure and reduce both storage and operational costs. We rebuilt the data warehouse using Big Query and dbt, deployed AI solutions for legal document screening, and built an autonomous chatbot to handle individual client case inquiries.

What this enabled

  • More than 90% reduction in database infrastructure costs after data warehouse rebuild
  • Around 70% cut in operational costs through custom internal AI solutions
  • Automated legal document screening and extraction using Google Vertex AI
  • Client-facing agentic chatbot reducing manual handling of case inquiries

05

Full-Circle Family

Family Tech · AI Product

Autonomous Agents
Fine-tuned
LLM for domain accuracy
WhatsApp
Direct user channel
AWS
Scalable microservices

Building an AI-powered mobile advisor for personalized parenting guidance

Full-Circle Family needed an accessible AI advisor that parents could reach through everyday channels. We led the architecture and implementation of an AI-powered chatbot accessible via WhatsApp, using a fine-tuned LLM combined with a domain-specific RAG knowledge base for accurate, safe parenting guidance.

What this enabled

  • Mobile-first AI advisor accessible directly through WhatsApp
  • Fine-tuned LLM with domain-specific knowledge base for consistent, accurate responses
  • Scalable microservices architecture on AWS for reliable production deployment
  • User behavior analytics dashboard to track engagement and inform product strategy

06

R-Cycle

Sustainability · Industrial Data Compliance

AI Workflow Systems
Multi-site
Manufacturer tracking
GCP
Cloud-native data layer
Compliant
Government mandate alignment

Giving manufacturers real-time visibility into recycling compliance across facilities

R-Cycle needed a centralized dashboard to help manufacturers monitor plastic production data and meet government recycling mandates. We led development of the customer-facing platform, rebuilt the data and file management layer on GCP, and introduced a comprehensive testing strategy for a previously untested application.

What this enabled

  • Cross-manufacturer tracking of plastic production data in a unified dashboard
  • GCP-based data transformation for scalable file and record management
  • Compliance visibility aligned with government recycling mandates
  • End-to-end testing strategy introduced to stabilize a previously untested codebase

07

Cinoware

Critical Infrastructure · Government Tech

AI Performance Monitoring
OpenCV
Navigation tracking
MLflow
Pipeline performance
Selenium
Automated test coverage

Connecting remote robot operation to ML performance monitoring in critical infrastructure

Cinoware builds testing and automation software for critical infrastructure in the government and telecommunications sector. We developed a remote robot operation console, implemented navigation tracking through image detection, and built an ML pipeline with performance monitoring to benchmark and improve algorithm accuracy over time.

What this enabled

  • Remote robot control application with command-issuance interface built in Python and React
  • Navigation tracking through real-time image detection using OpenCV
  • ML pipeline for algorithm benchmarking and performance tracking with MLflow
  • Selenium Grid testing suite with Allure reporting for systematic quality assurance

08

Applai

HR Tech · Agentic Workflow

Autonomous Agents
LangGraph
Agentic orchestration
Multi-LLM
Cover letter quality
LangSmith
Evaluation pipeline

Automating job application workflows with an agentic AI pipeline

Applai needed to accelerate the job search and cover letter process for users at scale. We built a fully agentic backend using LangGraph for workflow orchestration and multiple LLMs with RAG for high-quality, consistent cover letter generation, tracked and evaluated through LangSmith.

What this enabled

  • Agentic pipeline orchestrated with LangGraph for end-to-end job search automation
  • Multi-LLM cover letter generation with RAG for domain-specific accuracy and consistency
  • Selenium-based scraper backend built on FastAPI for live job market data
  • LangSmith integration for continuous model evaluation and quality tracking

09

kunveno

Enterprise SaaS · Microsoft 365

AI Workflow Systems
270M+
Teams users reached
Azure
Cloud architecture
OAuth
Org-wide authentication

Integrating an enterprise application directly into Microsoft Teams for 270 million users

kunveno needed to extend their SaaS product into the Microsoft Teams ecosystem to reach corporate users at scale. We architected the Azure cloud environment, implemented the Teams Message Extension, and built organization-wide authentication via Microsoft's delegated OAuth protocol.

What this enabled

  • Microsoft Teams Message Extension built to extend product reach to 270 million potential users
  • Microsoft Azure environment architected from the ground up as lead developer
  • Organization-wide authentication implemented via Azure AD and delegated OAuth
  • CI/CD pipeline managed across multiple deployment stages in close collaboration with the CTO