M. Shaharyar

> I build production AI systems — RAG pipelines, LLM integration, and semantic search that ship.

Solo engineer. I take AI products from schema to deployed, end to end: vector search over millions of chunks, multi-stage LLM pipelines, real-time infrastructure. Live products below — click them.

Highlights

Engineering focus

Built and operated production platforms end to end as a sole engineer

Designed real-time collaborative infrastructure using Operational Transformation

Built semantic search and content-analysis pipelines with Rust, Qdrant, PostgreSQL, and embeddings

Shipped multi-tenant SaaS workflows with authentication, audit trails, secure file handling, billing, and automation

Created OpenOT, an open-source Operational Transformation framework for collaborative editors

Selected Work

Engineering & Product

Recall

Flagship · Live RAG Product

Spoiler-aware RAG for fiction canon · live at recall.novusatlas.org

[TypeScript][Next.js][Qdrant][Qwen3-Embedding-8B][Multi-provider LLM (failover)][Radix][Tailwind]

A continuity-checking and canon research tool for fiction writers. Semantic search over millions of indexed text chunks with chapter-aware spoiler boundaries — queries only ever retrieve from canon the user has read.

Problem

Retrieval has to respect a per-user "read up to chapter N" cutoff — a query can never surface canon the reader hasn't reached yet.

Solution

Chunk-level provenance so retrieval respects the spoiler boundary; an embedding pipeline for long-form serialized fiction; credit-metered generation; full product built and deployed solo.

NOVA (Novus Atlas)

Real-Time Platform

Real-time collaborative writing platform · private beta

[Next.js][React][TypeScript][Rust][PostgreSQL][Redis][Qdrant][WebSockets][Socket.IO]

Collaborative fiction platform with an operational-transformation editor, custom Rust NLP engine, and vector-powered search. Built solo over two years.

Problem

Long-form collaborative fiction needs conflict-free concurrent editing, fast search across large manuscripts, and heavy NLP — all in one product.

Solution

OT convergence under concurrent editing; sub-100ms sync at hundreds of concurrent sessions; Rust services alongside a TypeScript product layer.

Impact & Stats

  • 1,000+ beta users
  • 500+ concurrent collaborators sustained at <80 ms p95 sync latency
  • 5M+ searchable content chunks indexed
  • Rust-powered semantic analysis and vector search infrastructure

OpenOT

Open Source

Open-source operational transformation framework · GitHub

[TypeScript][WebSockets][SSE][Operational Transformation][React][Lexical]

The OT engine extracted from NOVA, open-sourced. Framework-agnostic, with state-machine transforms, pluggable transports, storage adapters, and React/Lexical integration paths.

Kicklayer

Commercial SaaS

Client onboarding platform for agencies

[TypeScript][Next.js][PostgreSQL][AES-256-GCM][AI Workflows]

Multi-tenant client-onboarding SaaS — schema-driven intake, passwordless portals, secure credential exchange, and AI-assisted asset checks. Evidence of shipping breadth.

Services

What I take on

  • RAG systems: ingestion, chunking, embedding, retrieval, evaluation — from zero to production
  • LLM integration: multi-stage pipelines, structured output, streaming, provider failover
  • Semantic search over large or messy corpora
  • Real-time and collaborative features (OT, presence, sync)

Typical engagements: fixed-scope builds ($1.5k–8k) or ongoing hourly. I work async-friendly, overlapping US/EU mornings.

UI / UX

Interface craft & presence

Production-grade systems designed across NOVA, Kicklayer, and experimental tools. Each layout balances hierarchy, typography, and atmosphere to communicate product intent clearly.

Expertise

Technical proficiency & domain knowledge

Languages

  • TypeScript
  • JavaScript
  • Rust
  • Python
  • SQL

Frameworks

  • Next.js
  • React
  • Hono
  • Node.js
  • tRPC
  • oRPC

Databases

  • PostgreSQL
  • Qdrant
  • Redis
  • MySQL

Specializations

  • Real-time collaborative systems
  • Operational Transformation
  • Distributed synchronization
  • Semantic search and vector embeddings
  • AI-native workflows and LLM orchestration
  • Rust-powered NLP and content-analysis pipelines
  • Secure multi-tenant SaaS architecture
  • Product-focused full-stack engineering

Industries

  • AI Infrastructure
  • Collaborative Tools
  • Developer Tools
  • Workflow Automation
  • Content and Publishing Platforms
  • B2B SaaS

About

How I work

I'm a solo engineer. I build entire systems end to end — schema, backend, infrastructure, the model pipeline, and the interface a user actually touches — because I like owning the whole thing and I ship faster when nothing gets thrown over a wall.

Most of my work lives at the hard edges: retrieval that has to be correct, sync that has to converge, pipelines that have to stay cheap and fast in production. I care about systems that keep working after launch, not demos.

Outside the terminal I'm slowly building a JDM restomod — same instinct as the software: take something real, understand every part, and see the long project through.

Writing

Thoughts on design and semantics

Let’s collaborate

Available for contract engineering, advisory, and fractional CTO engagements

Reach out via email, schedule a call, or review the latest resume to see how I approach software delivery, contract execution, and technical leadership support.