Services for teams building with AI

LLM and chatbot evaluation

Your AI demos well, but nobody can say how often it’s wrong, or whether the last prompt change made it worse.

What I do

• Build a test set of real questions and expected behaviour for your product
• Measure hallucination, faithfulness, answer relevance, consistency and prompt quality with DeepEval and RAGAS
• Find failure patterns and fix the prompts, retrieval or guardrails behind them
• Run the evaluations automatically on every model, prompt or code change

You get: an evaluation suite, a baseline quality report, and pass/fail thresholds in your CI pipeline.

RAG systems

You want AI that answers from your own documents and data, without making things up.

What I do

• Design the ingestion, embedding and vector search pipeline, for example on MongoDB Atlas Vector Search
• Combine keyword (BM25) and semantic search, with query preprocessing and LLM reranking
• Measure retrieval quality and grounding, not just whether answers sound right
• Diagnose and fix an existing RAG system that gives weak or wrong answers

See it in practice: AI Test Case Generation with RAG

Autonomous QA and AI test automation

Your team spends too much time writing test scripts and fixing ones that break after every UI change.

What I do

• Turn user stories and Jira tickets into Gherkin scenarios and Playwright tests with LLMs
• Add AI-assisted self-healing for broken locators using Playwright MCP and live browser inspection
• Connect generation, execution and reporting to your CI/CD pipeline
• Keep humans reviewing what matters, so self-healing doesn’t hide real bugs

Track record: AI-enabled automation at SAP Ariba delivered ~60% less test authoring time and ~80% less regression effort.

Chatbots and conversational AI

You need an assistant that talks naturally with customers but never breaks your business rules.

What I do

• Build text and voice assistants, including phone calls through Twilio Voice
• Use the LLM for understanding while a deterministic state machine controls the flow
• Keep context across turns and complete real tasks such as bookings
• Write regression tests for multi-turn conversations, with turn-level logging to debug failures

See it in practice: Conversational AI Sales Platform, with 118 passing tests

AI applications, prototype to production

You have an AI idea or a prototype, and need it running securely for real users.

What I do

• Build full-stack AI apps with Python or TypeScript, FastAPI or Node.js, and React or Next.js
• Put deterministic safety rules beside the model where mistakes are costly
• Deploy on Azure or Oracle Cloud with Docker, Nginx, PM2 and HTTPS
• Set up tests and quality gates so the app keeps working after launch

See it in practice: VitalTriage and Orbit World Travels

AI quality strategy and team training

You’re putting AI into production and need a way to decide what’s good enough to ship, and teams who can check it.

What I do

• Set up the AI Quality Operating Model: risk-based strategy, measurable standards, automated systems and clear ownership
• Define release gates and quality thresholds for AI features
• Run hands-on workshops on LLM evaluation, RAG testing, Playwright with MCP and testing agents
• Coach QA engineers moving from script writing to evaluating AI

Background: led a team of 17 QA engineers at SAP Ariba and defined quality strategy for large-scale enterprise platforms.