Skip to content
View Ayaan2907's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report Ayaan2907

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Ayaan2907/README.md

Ayaan Kaifullah

Full-Stack AI/ML Engineer · New York, NY

LinkedIn npm WingMic Email

I build production AI systems end-to-end — agents, streaming APIs, infra, and UI. I own the stack, ship the features, and keep the platform running.


What I'm building

WingMic — you meet 20 people at a conference and forget them all by Monday. Speak 10–30s after a conversation; the LLM extracts names, companies, context, and follow-ups into a personal knowledge graph. Ask later: "who at Acme works on Rust?" — answer in <500ms.

Next.js 15 Bun Turborepo Drizzle + Turso Claude Sonnet OpenAI embeddings tRPC BetterAuth


Experience

AdvanceIQ.ai · Full-Stack AI/ML Engineer · Oct 2025 – Present · New York, NY

Sole engineer on a multi-product AI platform for VVIP finance clients. Own the full stack — architecture, infra, auth, streaming UI, background jobs, admin tooling, and every API route.

  • Power BI embed pipeline — Azure AD service principal auth, dual-method token generation with automatic fallback, Redis caching with proactive refresh; zero token-expiry incidents across all active sessions
  • ARIA chat — AI SDK v5 streaming over live dashboards; NL → DAX query translation via a semantic catalog of 894 measures and 283 bookmarks; tool routing across catalog lookup, client-side visual export, and conditional web search; per-user thread persistence, voice dictation, screen region capture
  • Admin platform — prompt sandbox with model tuning and full round-trip trace observability; user + report management; PostHog analytics; feature access control via Clerk metadata; global notification banners; DAX measure review interface
  • SRI Pricing module — server-side factor-rate engine (IRR + cash-on-cash matrices) with a three-layer assumption stack: user overrides → admin global → code defaults; Zod-validated, QA'd against source Excel
  • Discover — daily AI news platform on the same infra; Perplexity Sonar pipeline, Claude editorial selection, admin-controlled query lists and freshness windows, digest emails via QStash, custom story compose panel
  • Email + cron infra — QStash-backed async email queue with signature verification and automatic retry; Resend for delivery; per-user notification preferences; daily/weekly digest schedules

Next.js 15 React 19 TypeScript AI SDK v5 OpenRouter Claude Gemini Perplexity Sonar Power BI SDK PostgreSQL Redis QStash Prisma Clerk Vercel Sentry PostHog


Muvik · Forward Deployed Engineer (Founder-Level) · Jul – Oct 2025 · New York, NY

Built a dynamic text-to-workflow engine for logistics — plain English triggers composable workflows (billing, orders, scheduling, follow-ups, callbacks) constructed and executed at inference time. Adopted by 10+ companies in production. Flew to conferences, ran on-site onboarding, iterated from real operator feedback.

Next.js Mastra agents + workflows assistant-ui TypeScript


TinyCo.ai · Software Developer · Jun – Nov 2023 · Toronto, ON

Platform scaled 1 → 300K users. Replaced Flatfile with a custom CSV handler, saving $10K+/year. Owned CI/CD and release ops end-to-end — 99.9% uptime.

Python TypeScript React PostgreSQL GitHub Actions


Projects

Auto LLM Selector · npm · GitHub

Semantic classifier that scores prompts and routes them to the optimal LLM via OpenRouter. TensorFlow USE embeddings, multi-label classification, 85–92% cache hit rate at <200ms avg. 300+ weekly npm downloads.

TypeScript TensorFlow USE OpenRouter ESM


WingMic · In development

Voice-first networking memory. 10–30s of speech → structured knowledge graph → <500ms NL recall.

Next.js 15 Drizzle + Turso Claude Sonnet OpenAI embeddings tRPC


Video Analysis RAG · GitHub

End-to-end video ingestion pipeline — Whisper transcription + visual frame analysis in <60s per video. Vector search across 10K+ embeddings at <2s, 92% precision.

Python Pinecone Whisper multimodal RAG


Stack

LanguagesTypeScript Python JavaScript SQL C++

AI / AgentsMastra AI Claude GPT-4o Gemini Perplexity Sonar LangChain MCP Whisper TensorFlow Pinecone OpenRouter Vercel AI SDK v5

FrontendNext.js 15 React 19 Tailwind CSS assistant-ui SSE / streaming UI

BackendNode.js FastAPI PostgreSQL Redis QStash Drizzle Turso Prisma

InfraVercel Railway Kubernetes Sentry GitHub Actions Clerk Docker


Education

Monroe University — M.S. Computer Science · Expected Jul 2026 · New York, NY University of Michigan — M.S. Computer Science · Graduate coursework, Sep – Dec 2024

Pinned Loading

  1. Multimodal-RAG-video-analysis Multimodal-RAG-video-analysis Public

    Description, README in progress

    TypeScript

  2. text-completion-AI-extension text-completion-AI-extension Public

    TypeScript

  3. simstudioai/sim simstudioai/sim Public

    Build, deploy, and orchestrate AI agents. Sim is the central intelligence layer for your AI workforce.

    TypeScript 29k 3.7k

  4. advanced-ai-code-editor advanced-ai-code-editor Public

    Forked from judge0/ide

    Judhe0 based code editor with AI capibilities.

    JavaScript 2

  5. Visual-Attendence-Tracking Visual-Attendence-Tracking Public

    Python