Founder · Paiteq

Navin Sharma

Founder of Paiteq (legal entity: Paiteq Private Limited). AI engineering operator since 2017. Today the team ships production AI for regulated buyers across healthcare, fintech, insurance, and logistics.

What I do

I run Paiteq, a Bengaluru- and Dallas-based AI engineering studio. Daily Claude Code and OpenAI Codex operator. We ship production AI agents, RAG systems, voice agents, and intelligent document processing for clients across six industries. Eval-first delivery, model-agnostic across Claude, OpenAI, and open-source. Audit-logged, regulator-defensible.

How Paiteq started

Paiteq is the AI engineering arm of the team behind GetWidget, the open-source Flutter UI library shipped since 2019. The getwidget repo sits at 4,800+ GitHub stars (2026-Q2), with corresponding packages on pub.dev used in production by Flutter teams worldwide. The Flutter dev shop arm of the same group operates as HireFlutterDev. Eight years of shipping production mobile gave us the operational discipline. In 2024 we pivoted the studio to AI engineering. Same legal entity (Paiteq Private Limited), same engineers, same eval-first instinct, new problem space.

Today Paiteq runs 12 service practices (AI consulting, AI agents, RAG, LLM apps, MLOps, chatbots, AI integration, AI migration, workflow automation, generative AI, machine-learning development, RPA) and 6 industry verticals (ecommerce, fintech, healthcare, insurance, logistics, SaaS).

On the operator side I run two production properties: Aerostack (developer platform on Cloudflare's edge, 100+ active teams, 50+ agents shipped, 2026-Q2) and Nyburs (India hyperlocal social network, 500K+ users, 12+ Indian languages). Running these every day is where the AI-delivery muscle gets exercised — failure modes only show up when the on-call rotation is yours.

What I write about

Long-form essays on AI delivery, LLM evaluation methodology, RAG architecture, voice-agent latency budgets, agent orchestration with LangGraph and CrewAI, and regulated-buyer deployment under HIPAA, SR 11-7, FFIEC, and ISO 42001.

Recent posts live on the Paiteq engineering blog. Pillar pages worth bookmarking: AI agent development company, RAG development services, LLM development services, AI consulting services, chatbot development services, generative AI development services, and machine learning development services.

Recent operator benchmarks

Numbers I quote in talks and on first calls, with dates so you can check them yourself:

  • RAG citation gate on a 1,840-document clinical-knowledge corpus: recall@5 0.88, p95 retrieval latency 240ms (Ragas 0.2.x, 2026-Q1, $14 total Claude API spend on the eval run).
  • Multi-provider router fallback: 99.4% successful completion across a 30-day window spanning two Anthropic and one OpenAI outage (2026-Q1, in-house gateway, 1.2M requests).
  • Voice agent first-token latency: sub-380ms p95 on OpenAI Realtime API with a 2-node WebRTC POP layout (Bengaluru + Dallas, 2026-Q2).
  • GetWidget repo: 4,800+ stars on github.com/ionicfirebaseapp/getwidget as of 2026-Q2; corresponding pub.dev packages used in production by Flutter teams worldwide.

Across the Paiteq engineering practice the same eval discipline applies to every workload, eval set written in week two, thresholds turn green before production wire-up.

What I will and won't take on

Yes: production AI workflows in 4 to 9 weeks with named eval criteria, regulated-buyer deployments (HIPAA, FFIEC, SR 11-7), model-agnostic engineering across Claude, OpenAI, and open-source, voice and document automation, RAG pipelines that pass citation gates.

No: 50+ headcount ML team work (Big-4 territory), single-vendor SaaS resale, pure research, demos without measurable workflow value, single-model vendor lock-in.

Where to find me