~/sadiq.khan▍
Available
● status: available/location: remote, global

I build AI-native products
end-to-end
from model to mobile,
from API to App Store.

I'm a full-stack engineer with 2+ years shipping production AI products, mobile apps, and scalable web platforms. I specialize in LLM integration, RAG systems, and turning ambiguous ideas into shipped software.

See shipped workStart a conversation
0+
years shipping
0+
products deployed
0M+
LLM calls / day
0< h
response time
01 / About

I write code that ships — and stays shipped.

I started writing code at 25 because I wanted to build a Discord bot. That bot is long dead, but the instinct it gave me — figure it out, ship it, iterate — has shaped everything since.

What I do best sits at an intersection most engineers avoid. I can architect a Next.js frontend, design the Postgres schema behind it, deploy the Node API on AWS, wrap the whole thing in a React Native app, and bolt on an LLM-powered feature that actually moves a business metric. That combination is rare because it's hard. Most people pick a layer. I learned the whole stack because the products I wanted to build demanded it.

My edge is treating AI as a first-class engineering discipline, not a magic layer. I've built RAG pipelines that reduced hallucination by 60%, agent systems that handle multi-step workflows autonomously, and orchestration layers that cut LLM costs by 40% through caching and routing. I don't prompt-and-pray. I measure evals, track token economics, and design fallbacks.

I work best with founders and product teams who already know what they're solving but need someone who can translate that into a system. I'm not the person for “make it pop” or “add a chatbot.” I'm the person you bring in when the system is breaking at scale, when the AI feature isn't performing, or when the mobile app needs to ship in eight weeks and there's no room for guesswork.

02 / Stack

The tools I reach for, and why.

Stack lists are noise. What matters is knowing when to use what, and what to avoid. Here's how I think about the layers I work in every day.

Signature skills

01

AI-Native Product Engineering

Not 'adding AI' to existing products, but building products where LLMs are core to the value proposition. I design the data flows, eval systems, and fallback paths.

02

Full-Stack Mobile + Web Cohesion

One engineer, one mental model, ship to web and mobile simultaneously without duplicating effort or reasoning.

03

Production-Grade RAG Systems

Retrieval that doesn't hallucinate, scales beyond demos, and is measurable from day one.

▍Frontend

React / Next.js

Server components, app router, edge functions. My default for any web product that needs to rank, convert, and load fast.

TypeScript

Non-negotiable on anything beyond a prototype. Catches the bugs I'd otherwise ship at 2am.

Tailwind CSS / CSS Modules

Utility-first for speed; modules when design systems demand scoped styling.

Framer Motion

Micro-interactions that make products feel finished, not merely functional.

▍Backend

Node.js (Fastify, Express)

Fastify for new APIs, Express when integrating with legacy. Both well.

Python (FastAPI)

Default for anything ML-adjacent. Async, typed, fast to ship.

PostgreSQL

My default relational store. I write migrations like I write code: carefully and reversibly.

Redis / BullMQ

Background jobs, caching, rate limiting — the unglamorous infrastructure that keeps apps alive at 3am.

▍Mobile

React Native

Ship iOS and Android from one codebase, OTA updates, EAS Build. My mobile default.

▍AI / ML

LLM Integration

OpenAI, Anthropic, OSS via vLLM / Ollama. API design, streaming, function calling, structured outputs, cost modeling.

RAG Pipelines

Vector stores (Pinecone, pgvector, Qdrant), chunking, hybrid search, reranking. Shipped RAG handling 10k+ queries/day.

AI Agents

LangGraph and custom orchestrators. Multi-step reasoning, tool use, memory, human-in-the-loop patterns.

▍DevOps / Cloud

AWS (ECS, Lambda, S3, RDS)

My default cloud. I think in terms of cost per request, not cost per month.

GCP (Cloud Run, Vertex AI)

Reach for it when AI workloads demand managed GPU inference.

Docker

Every project runs in containers. Every one.

CI/CD

GitHub Actions, Vercel, EAS — automated tests, preview deploys, mobile OTA pipelines.

03 / Projects

Four products. Real metrics. Real lessons.

Not tutorials, not clone apps. Each of these shipped to real users and either made money, saved time, or both.

01 /AI SaaS

Atlas

AI-powered research synthesis for legal teams.

Problem

Mid-size law firms spend 60% of billable hours on document review. Existing legal tech is search-first; lawyers still have to read everything and synthesize it themselves.

Stack
Next.js 14TypeScriptPython / FastAPIPostgreSQL + pgvectorOpenAI GPT-4LangChainAWS ECSRedisStripe
Key engineering decisions
  • ▸Switched from pure semantic search to hybrid BM25 + vector retrieval — moved answer accuracy from 68% to 91%.
  • ▸Built a custom cross-encoder reranking layer on top of retrieved chunks; out-of-the-box wasn't good enough for legal use cases.
  • ▸Designed the UX around citation anchors, not chat bubbles — every answer is traceable back to source paragraphs. Trust lives in traceability.
Impact

Beta with 3 mid-size law firms. Average document review time dropped from 4.2 hours to 38 minutes (~85% reduction). 12 paying customers at $2k/mo within 60 days of launch.

Lessons

In regulated domains, citations matter more than confidence. The 'AI part' was 30% of the work; the other 70% was auth, billing, document ingestion, and the unglamorous plumbing.

02 /Web App

Pulse

Real-time team health dashboard for distributed engineering managers.

Problem

Engineering managers of remote teams had no signal on team health beyond GitHub commit counts. Engagement surveys ran quarterly and were always lagging the actual problem.

Stack
Next.jsTypeScripttRPCPrismaPostgreSQLRedisTremorVercelGitHub + Slack APIs
Key engineering decisions
  • ▸Built an event-sourced data model so we could replay team metrics and backtest new anomaly-detection algorithms against historical data.
  • ▸Pulled signals from GitHub PR cycle time, anonymized Slack activity, and a 2-question weekly pulse survey — three independent sources, not one.
  • ▸Surfaced anomalies, not dashboards — managers got pinged only when patterns deviated from baseline. Dashboards are noise; alerts are signal.
Impact

40+ engineering teams in private beta. Teams using Pulse for 3+ months reported 31% improvement in identifying at-risk projects early. 92% weekly-active retention among managers.

Lessons

Anonymity is non-negotiable for adoption. We initially showed individual contributor signals and got immediate pushback; pivoting to team-aggregated metrics unlocked usage. Weekly pulse surveys are the sweet spot — daily is noise, monthly is stale.

03 /Mobile App

Reps

AI-powered sales call coaching for field sales teams.

Problem

Field sales reps get coaching in quarterly reviews, not after the call. Existing call recording tools transcribe but don't actually coach.

Stack
React Native (Expo)TypeScriptNode.js / FastifyPostgreSQLSelf-hosted WhisperGPT-4AWS S3 + CloudFrontEAS Build
Key engineering decisions
  • ▸On-device audio capture with chunked uploads to handle unreliable field connectivity — reps lose signal in parking garages and basements constantly.
  • ▸Built an async pipeline: upload → transcribe → analyze → notify. Reps got feedback in under 5 minutes, not next quarter.
  • ▸Coaching prompts are role-specific (SDR vs AE vs CSM). Generic feedback doesn't land with anyone.
Impact

Pilot with 80 reps at a B2B SaaS company. New AE ramp time dropped from 5.2 months to 3.8 months. Reps using the app closed 22% more deals in their first 90 days.

Lessons

Mobile field conditions are brutal. Battery, network drops, OS background restrictions — we rewrote the upload system twice. Coaching tone matters more than content: blunt critiques got ignored; softened, specific feedback got 3x more engagement.

04 /API / Backend

Relay

Multi-model LLM orchestration API for enterprise teams.

Problem

Enterprises wanted LLM features but faced vendor lock-in, unpredictable costs, and compliance gaps. Building this in-house took 6+ months of dedicated work.

Stack
Python / FastAPIPostgreSQL + pgvectorRedisKubernetes (EKS)LiteLLMOpenAI + Anthropic + OSS modelsPrometheus / GrafanaTerraform
Key engineering decisions
  • ▸Built a unified interface abstracting model providers — clients swap between GPT-4, Claude, and self-hosted models with a config change, no code change.
  • ▸Implemented semantic caching at the prompt level, cutting repeat LLM costs by ~40% without measurable quality loss.
  • ▸Added automatic fallback chains: if GPT-4 rate-limits, fall back to Claude; if both fail, return cached or graceful error. Designed for graceful degradation from day one.
  • ▸Per-tenant cost tracking and rate limiting at the edge — finance teams get itemized bills, abuse is caught before the invoice.
Impact

Handles ~2M LLM calls/day. Powers AI features for 8 enterprise clients across fintech, healthcare, and legal. 99.94% uptime over 6 months. Average cost per request reduced 40% vs. direct OpenAI usage.

Lessons

LLM APIs are not normal APIs. Latency variance is massive, rate limits are real, and providers deprecate models without warning. Building for failure was the best decision I made. Enterprise sales needs SSO, audit logs, and a SOC 2 story before they even look at the product.

04 / Services

What I take on, and how it usually goes.

Outcome-focused engagements. No retainers, no mystery. You know what you're getting and what it costs before we start.

01

AI-Powered App Development

I design and ship products where LLMs are core to the value — not bolted on as a feature, but engineered into the data flow, evaluation pipeline, and user experience. You get a working system with measurable AI performance, not a demo that breaks the moment real users show up. Typical engagements: RAG systems, AI agents, multi-model orchestration, and AI-native SaaS products from zero to revenue.

02

Full-Stack Web Development

I build production web applications end-to-end — frontend, backend, database, deployment — with a senior engineer writing every layer, not a team passing the work down a chain. Modern stacks (Next.js, TypeScript, Postgres), clean architecture, and code you'd be comfortable handing to your next hire. Typical engagements: MVPs in 8–12 weeks, complex feature builds, and codebase rescues when the original team didn't ship.

03

Mobile App Development

Cross-platform iOS and Android apps shipped from a single codebase, with native-quality UX and OTA updates that don't require App Store review. I work in React Native / Expo by default, Flutter when the project demands it, and I don't disappear after launch — I'll set up your CI/CD, crash reporting, analytics, and release pipeline. Typical engagements: consumer apps, internal tools, and field-force apps where connectivity is unreliable.

04

API Design & Backend Architecture

I design APIs that are easy to consume, hard to misuse, and built to scale beyond your first thousand users. REST, tRPC, or GraphQL — whatever fits the use case — with proper auth, rate limiting, observability, and documentation that is actually useful. Typical engagements: greenfield API systems, third-party integrations, and refactoring monoliths that have outgrown their original design.

05

Technical Consulting & Code Review

Bring me in when something is broken, slow, or about to break, and you need a senior engineer's eyes without a full-time hire. I do architecture reviews, performance audits, AI system evaluations, and code reviews with written reports and concrete recommendations — not vague advice. Typical engagements: 2–4 week sprints, pre-launch audits, and 'we think this is the problem but we need a second opinion' calls.

05 / Testimonials

Words from people who hired me.

Migrated our LLM infrastructure in six weeks. Zero downtime, 40% cost reduction, and the new system actually has observability — which is more than I can say for what we had. If you need someone who understands both the AI layer and the production layer, hire this person. I've worked with a lot of 'AI engineers' who can prompt. This one can ship.
Marcus Chen
Staff Engineer, FinPath AI
We needed to ship an AI feature our enterprise clients were asking for, and our roadmap was already full. Sadiq came in, scoped it in a week, shipped a working RAG system in six weeks, and our largest client renewed their $400k contract specifically because of it. The ROI math was obvious within the first month.
Priya Sharma
VP of Product, Ledgerline
I talked to seven engineers before this one. Six wanted to rebuild everything from scratch. Sadiq spent the first call asking why we built what we built, then proposed a fix that didn't require a rewrite. That's when I knew I found the right person. Shipped our mobile app in nine weeks, 4.8 stars on the App Store, and I've rehired him twice since.
Daniel Okafor
Founder & CEO, Reps
06 / Contact

Let's talk about
what you're building.

The fastest way to reach me is hello@sadiqkhan.dev. I read everything and reply within 24 hours on weekdays.

Send a briefReview the work first
Response
Within 24 hours
Capacity
2–3 new projects
Window
Q4 2026 onward

If I can't help, I'll point you to someone who can. That's the deal.