Available for freelance projects

Reliable AI automation, monitoring, and document intelligence - built to survive real-world failure, not just demo well. Three things you can try right now: an AI assistant for university students, an OCR engine that pulls text out of scanned documents, and a tool that poisons code against AI training scrapers.

UET GPT - open app GitHub
6 shipped systems664-test OCR suite299-test monitoring suiteRAG + document intelligenceOpen for contract hire →

What I build

Hiring Guide & Rates →

Shipped projects

LIVEFeatured

UET GPT

AI assistant for university students

A ChatGPT-style assistant that answers UET students' questions (admissions, fees, courses) using real university data instead of guessing.

Live at uet-gpt.vercel.app - sign-in, rate limiting, error monitoring, automated tests.

Next.jsTypeScriptConvexRAGVercel AI SDKClerk
About

I'm Ibrahim Salman - I build AI assistants, OCR pipelines, and monitoring/automation systems designed to survive real-world failure, not just demo well.

More about me →

Availability
Available for freelance projects

AI chatbots (RAG) · monitoring & extraction · OCR · automation - Python, TypeScript, Java.

B.L.A.S.T.

LIVE

Reads text out of documents (OCR)

Drop in a PDF, PowerPoint, or image and it extracts the text into clean Markdown/Word files: measured 61.6% CER reduction (fewer errors to proofread by hand) and 3.9x faster than baseline OCR on a 14-page benchmark corpus.

61.6% CER reduction, 3.9x faster - fewer OCR errors, less manual cleanup, measured on a 14-page benchmark corpus.

Marketplace Monitoring & Alerting

COMPLETE

Tracks marketplaces, alerts you the moment something changes

Watches job and freelance marketplaces for new postings and sends a structured email digest the moment one appears: fail-closed on every external dependency so an outage never means a silently missed lead, backed by 299 tests including fuzz and adversarial suites.

Menu Data Pipeline

COMPLETE

17 sources -> validated data -> live spreadsheet

Concurrently pulls menu data from 17 restaurant sites, validates and cleans it with Pydantic and pandas, and pushes a unified dataset straight into a live Google Sheet.

Nightshade

RELEASED

Poisons code against AI training scrapers

A tool that rewrites your code so it still compiles and runs correctly for humans and compilers, but is designed to degrade as AI training data.

Quantum Singularity

LIVE

WebGL particle art you control with your hands and voice

A real-time WebGL2 particle simulation (up to 110,000 GPU particles) driven by hand gestures and live audio, running entirely in the browser from a single static HTML file.

Assistant

Ask my AI assistant anything

A live demo of my LLM work - ask about projects, stack, and availability.

Philosophy
"Make it work. Prove it works. Make it survive production."
Have a project?

Documents to process, a source to monitor, or a workflow to automate?

Hire Ibrahim / Scope Project

How I work

01

Understand

Problem, constraints, and edge cases before any code.

02

Design

Scope the architecture, pick failure modes to design against.

03

Build

A working vertical slice, not scaffolding for its own sake.

04

Harden

Tests, retries, timeouts, and observability - before it ships.

05

Ship

Deployed, documented, and monitored in production.

Frequently asked questions

Is Ibrahim Salman available for freelance projects and hiring?

Yes. Ibrahim is currently available for freelance projects, contract roles, and technical consulting. He replies within 24 hours to project inquiries. Based in Islamabad, Pakistan (UTC+5), he works remotely with clients worldwide across North America, Europe, and Asia.

Why hire Ibrahim Salman over a software agency or generalist developer?

Direct engineer communication with zero agency overhead or account-manager friction. Ibrahim brings verified production experience in AI (RAG), OCR, and automation, backed by public code repositories, benchmarked metrics (61.6% CER reduction, 3.9x speedup in B.L.A.S.T.), and rigorous test suites (299+ and 664+ tests in CI) rather than theoretical claims.

What kind of projects does Ibrahim Salman build?

Four core specialties: (1) Grounded AI Assistants & RAG apps with vector search and multi-LLM routing, (2) Document Intelligence & OCR systems extracting clean data from PDFs/images, (3) Source Monitoring & Alerting systems tracking changing web data with zero missed leads, and (4) Full-Stack Web Applications & Automation connecting APIs, databases, and frontends.

What technologies and tech stack does Ibrahim Salman use?

Frontend & Full Stack: Next.js 15 (App Router), React 19, TypeScript, Tailwind CSS, Framer Motion. Backend & Data: Python 3.12, FastAPI, Convex, SQLite, PostgreSQL, Pandas, Pydantic. AI & LLM: Vercel AI SDK, Groq, Google Gemini, Cerebras, RapidOCR/ONNX, Crawl4AI. Testing: Playwright, Pytest, Vitest.

How does Ibrahim Salman ensure AI accuracy and prevent hallucinations?

By designing strict retrieval-augmented generation (RAG) architectures: responses are strictly grounded in retrieved and chunked vector data, multi-provider LLM fallbacks guarantee uptime, and the assistant explicitly admits when data is unavailable rather than fabricating answers.

How do I start a project or hire Ibrahim Salman?

Email ibrahim.pk848@gmail.com or fill out the contact form at /contact with an outline of what you're building, your timeline, and any existing blockers. Ibrahim will reply within 24 hours to schedule a kickoff discussion.

How is pricing and project scoping determined?

Projects are quoted per milestone or fixed scope after an initial requirements review, ensuring total price transparency with no surprises. See /pricing.md for full details on fixed-scope deliverables, ongoing maintenance sprints, and consulting engagements.

Are the portfolio projects real, verifiable working software?

Yes. UET GPT is an active AI assistant in production at uet-gpt.vercel.app (sign-in required to use it), B.L.A.S.T. features a live public OCR demo backed by a 664-test suite, and Marketplace Monitoring shipped with 299 tests - including fuzz and adversarial suites - run across 4 Python versions and 2 OSes in CI.