Vinay C

Staff Software Engineer — Generative & Agentic AI: evaluation, RAG, multi-agent systems

Bengaluru, India · vinayatpesit@gmail.com · vinayc.me · linkedin.com/in/imvinayc · github.com/vinz

Summary

Staff engineer building production GenAI systems: evaluation, RAG and agents. Built the agentic voice assistant that replaced a rule-engine support IVR (6–7M calls a year) in under three months, cutting transfer rate from ~28% to under 20%. 14+ years across Intuit, Microsoft, Docusign, Oracle (Principal MTS) and SAP; mentors senior engineers; granted patent holder. Open to Staff and Principal GenAI engineering roles.

Experience

Intuit — Staff Software Engineer

Jan 2026 – Present · Bengaluru

As the Staff engineer on a team of six, built the agentic voice assistant for a customer-support IVR handling 6–7M calls a year, replacing a rule engine with static responses. Concept, design, evaluation and production launch in under three months.
Evaluated voice platforms (NLX, Amazon Nova Sonic, Google GECX) on 20+ parameters (voice quality, latency, fit with the existing landscape, integration complexity, observability, agent architecture and ReAct support, model flexibility, harness, guardrails); NLX was chosen.
RAG-first design for grounded answers over 10,000+ editor-maintained articles (hundreds updated daily): hybrid search and reranking, with the chunking strategy chosen through repeated evaluated experiments. When it cannot resolve, it captures a structured problem summary and routes to one of 200+ expert queues by intent, subscription and product.
Results: transfer rate down from ~28% to under 20%, and speech-to-text word error rate from 15% to 7%.
Evaluation at every layer: word error rate for speech-to-text and text-to-speech, retrieval and ranker metrics, Ragas for generation, plus completeness, helpfulness and accuracy of responses. Speech errors propagate into every downstream metric.
Built an in-house call simulator and evaluator that scores calls across many parameters, in a few weeks. A second team building in-app voice support now reuses it, with audio arriving over RTC from the product instead of the phone.
Guardrails for a financial-advice setting: documented boundaries on what the assistant may and may not say and how it handles ambiguity, kept honest with red-teaming exercises and automated checks.
Engineered latency across the whole voice path: model choice, prompt caching, article caching and pre-generated answers.
Own regression evaluation for the GenAI/RAG stack: golden dataset from real customer utterances, daily automated runs over 4M+ datapoints a year, leadership-facing reporting. Led embedding-model evaluation (BGE-M3, Stella 1.5B) and shipped per-query explainability (MRR, mAP, Hit Rate) for non-ML stakeholders.
Security Pod: ~21 vulnerability tickets closed via AI-assisted remediation, 55+ PRs for human review, adopted by the team. Built a reusable agentic debugging skill and championed agentic coding workflows.

Microsoft — Senior Software Engineer

Dec 2024 – Dec 2025 · Bengaluru

On the five-engineer Power Pages team reinventing a flagship product from drag-and-drop to natural-language interaction across design, development, publishing and production maintenance. Led a set of features with a few engineers.
Built features of a multi-agent system that creates, modifies and manages web applications for enterprise, sovereign and government-grade customers, with security and data sovereignty built in. Part of the Power Platform (25M+ monthly active users). Details confidential.

Docusign — Senior Software Engineer

Apr 2024 – Dec 2024 · Bengaluru

Built SDK platform support across 40+ open-source packages maintained by Docusign, via CodeGen SDK generation.
Reduced open customer GitHub issues by ~40% in five months.
Company hackathon finalist — Azure OpenAI tool helping developers debug Docusign integrations.

Oracle — Principal Member of Technical Staff

Feb 2022 – Mar 2024 · Bengaluru

Headed modernization of Enterprise Manager — a 30-year product with 10,000+ live customers — onto current technologies without breaking existing integrations.
Post-renovation: ~20% higher user engagement, ~40% better system efficiency, ~25% higher customer satisfaction.
Shaped design, strategy, and roadmap with senior directors.

SAP Labs — SWE → Senior SWE → Associate Architect

Jul 2012 – Jan 2022 · Bengaluru

Architected cloud-native microservices on Cloud Foundry for multi-cloud deployment; led design of a cloud solution improving scalability ~50% and cutting infra costs ~40%.
Created a Data Privacy & Protection microservice that led to a granted patent and improved EU-regulation compliance across products.
Led a team of 8 delivering data-enrichment solutions, presented at SAP TechEd (leads from Microsoft and Sony); trained 200+ colleagues on UI5 / Fiori.

Leadership & Influence

Mentor senior software engineers at Intuit.
Interviewer on 30+ interview loops at Intuit and 10–15 at Microsoft, mostly system design rounds.
Cross-team reuse: the call simulator and evaluator now serves a second team, and AI-assisted remediation was adopted by the Security Pod.

Skills

Evaluation & RAG: RAG, hybrid search, reranking, chunking strategies, embeddings, FAISS, golden datasets, Ragas, retrieval and generation metrics (MRR, mAP, hit rate), speech (STT/TTS) evaluation, latency optimisation, prompt caching, call simulation, vendor evaluation

Agents: Generative & agentic AI, multi-agent systems, LLMs, LangChain/LangGraph, MCP, agentic coding workflows

Platform: Azure, AWS, Kubernetes, Docker, serverless, Cloud Foundry, microservices, event-driven design, PostgreSQL, Redis, RabbitMQ, SAP HANA

Languages & frameworks: Java, JavaScript/TypeScript, Node.js, Spring Boot

Patent & Recognition

Granted patent: Providing Data Protection and Privacy as a Service in a Cloud Platform
Runner-up, SAP Global Intrapreneurship Challenge (600+ entries)
Speaker — SAP TechEd and industry forums

Education

B.E. Computer Science — PES University, Bengaluru (2009–2012)
Diploma, Computer Science — PES University (2006–2009)