01

Voice assistant for customer support

Intuit · Staff SWE · Team of 6 engineers
Agentic AI · RAG · Speech · Evaluation · Routing

Problem

The support IVR handles 6–7 million calls a year. It ran on a rule engine with static responses: callers got a fixed answer or an article link, and calls routed to the wrong expert were transferred again, wasting expert time.

Approach

Replaced it with an agentic system on LLMs and RAG, grounded in 10,000+ editor-maintained articles, hundreds of which change every day. It holds a real conversation: asks follow-ups, works through the caller's problem and answers from current content. When it can't resolve the issue, it captures the problem and its classification, then routes to one of 200+ expert queues using intent, subscription and product. Concept to production launch in under three months.

Key decisions

  • Choosing the platform by evidence. Before settling on the voice platform (NLX), evaluated Amazon Nova Sonic and Google GECX too, scoring each on 20+ parameters: voice quality, latency, compatibility with the existing landscape, integration complexity, observability, support for ReAct-style interaction with models and the agent architecture around it, support for different models, the agent harness and guardrails.
  • Grounding first. A RAG pipeline with hybrid search and reranking, so answers come from current articles and not from the model's memory.
  • Chunking by experiment. Many chunking strategies, each scored against the retrieval evals, so the choice was measured and not assumed.
  • Evaluation at every layer. Word error rate for speech-to-text and text-to-speech quality, retrieval and ranker metrics, Ragas for generation, and scores for completeness, helpfulness and accuracy. Speech errors flow into every downstream metric, so each layer had its own gate.
  • Our own test bench, reused. That evaluation work led to an in-house call simulator and evaluator that scores calls across many parameters, built in a few weeks. A second team building voice support inside the product now uses the same evaluator, with audio arriving over RTC from the app instead of the phone.
  • Guardrails for financial advice. The assistant works in a financial product, so what it may and may not say, and how it handles ambiguity, is documented and enforced with red-teaming exercises and automated checks.
  • Latency as a requirement. Optimised the whole voice path: the model, prompt caching, article caching and pre-generated answers.

Impact

~28% → <20%

Transfer rate

15% → 7%

Speech-to-text word error rate

6–7M

Calls / yr

02

GenAI regression evaluation

Intuit · Staff SWE · 2026
RAG · Evals · Embeddings · Explainability

Problem

A production GenAI / RAG stack was shipping changes with no reliable way to know whether retrieval quality had regressed for real customers.

Approach

Built a golden dataset grounded in real customer utterances and automated daily regression runs. Led embedding-model evaluation (BGE-M3, Stella 1.5B) for the retrieval path, and shipped per-utterance explainability — MRR, mAP, Hit-Rate on every query — so non-ML stakeholders can debug a bad score without filing a ticket.

Impact

4M+

Datapoints evaluated / yr

Daily

Automated regression runs

03

AI-assisted security remediation

Intuit · Staff SWE · 2026
Agentic coding · Skills · Security

Problem

A backlog of vulnerability tickets competed with feature work, and each fix was small, repetitive and easy to get subtly wrong.

Approach

Drove AI tooling on the Security Pod: agent-assisted remediation with every change reviewed by a human, plus a reusable agentic debugging skill other teams can adopt.

Impact

~21

Vulnerability tickets closed

55+

PRs generated for review

04

Reinventing Power Pages with agents

Microsoft · Senior SWE · 2024–25 · Team of 5 engineers; led a feature area
Multi-agent systems · Power Pages · Enterprise & sovereign

Problem

Power Pages, a flagship Microsoft product, was built around drag-and-drop authoring. The goal was to reinvent the whole experience around natural language, from design and development to publishing and running a site in production, for enterprise, sovereign and government-grade customers.

Approach

Led a set of features, with a few other engineers, on a five-engineer team building a multi-agent system: several agents that create, modify and manage web applications from conversation. Security and data sovereignty requirements had to hold at that scale, because the customers are regulated and government organisations. Much of the detail remains confidential.

Impact

Sovereign

Enterprise and government-grade customers

Multi-agent

Create, modify, manage web apps

05

Data Protection as a Service

SAP Labs · Associate Architect · 2015–22
Privacy-by-design · Microservices · Cloud Foundry

Problem

Cloud applications needed data protection & privacy as a first-class, reusable capability — not something each team reinvented under deadline.

Approach

Designed and built a data-protection-and-privacy microservice on the cloud platform, packaged so any application could consume it — the architecture that became a granted patent. In the same period, architected cloud-native microservices on Cloud Foundry for multi-cloud deployment.

Impact

1

Granted patent

−40%

Infra cost

+50%

System scalability

06

Modernizing Enterprise Manager

Oracle · Principal MTS · 2022–24
Platform modernization · UX strategy

Problem

A 30-year-old product with 10K+ live customers had accumulated the drag of three decades — slow, dated, and hard to love.

Approach

Headed the modernization effort and shaped design, strategy and roadmap directly with Senior Directors — treating a legacy renovation as a product problem, not just a rewrite.

Impact

+20%

User engagement

+40%

System efficiency

+25%

Customer satisfaction

07

SDK platform at Docusign

Docusign · Senior SWE · 2024
Code generation · Developer experience · Open source

Problem

Dozens of hand-maintained open-source SDK packages drifted from the API and piled up customer GitHub issues.

Approach

Built SDK platform support across 40+ packages through CodeGen SDK generation, so fixes land once and ship everywhere. Also a company hackathon finalist with an Azure OpenAI tool that helps developers debug Docusign integrations.

Impact

40+

Packages on one platform

−40%

Open customer issues, in 5 months

Want the detail behind any of these?

Happy to walk through the design decisions and numbers. The résumé has the full history.

Message me on LinkedIn →