Ambreen Habib

Full-Stack & ML Engineer specializing in RAG, LLM Evals, and AI-driven development.

Based in: Lahore, PakistanMost recently: Software Engineer, Spacebar TechnologiesExperience: 1 year

Ambreen is a high-achieving Computer Science graduate with a 3.95 CGPA who builds and rigorously evaluates end-to-end AI products using Python, React, and TypeScript. She has a proven track record of shipping production-ready applications, from fine-tuned BART models for legal summarization to RAG assistants with comprehensive automated testing suites.

Experience

Software Engineer

Spacebar TechnologiesLahore, Pakistan (Onsite)workNov 2025 – Jun 2026

Summary

Owned client projects from requirements to production with React.js, Next.js, TypeScript, and Supabase, communicating directly with clients. Built RESTful APIs, authentication, and RLS-secured PostgreSQL schemas, plus real-time admin dashboards. Used Claude Code and GitHub Copilot to prototype, ship, and review code faster, testing all generated code before release.

What I did

Contributed across frontend, backend, database, authentication, and real-time features.

Results

Replaced manual or disconnected workflows with centralized applications that made day-to-day processes more organized and efficient.
Built and delivered multiple real client-facing internal applications from scratch to production in approximately 2 months.
Supported internal client applications typically used by around 10-15 people.

React.jsNext.jsTypeScriptSupabaseRESTful APIsPostgreSQLClaude CodeGitHub Copilot

Software Engineer Intern

Spacebar TechnologiesLahore, Pakistan (Onsite)internshipSep 2025 – Oct 2025

Built an admin course-enrollment platform from scratch with React.js and Supabase, including real-time chat.

React.jsSupabaseReal-time chat

MERN Stack Developer Intern

10PearlsLahore, Pakistan (Remote)internshipJun 2025 – Aug 2025

Built a secure notes app (React.js, Express.js, PostgreSQL) with JWT refresh tokens, rate limiting, and API tests.

React.jsExpress.jsPostgreSQLJWTAPI tests

Skills

Technical

Evals

RAG

React.js

Claude Code

Supabase

Python

Next.js

GitHub Copilot

pgvector

FastAPI

TypeScript

Tailwind CSS

Deep Learning

Fine-tuning

Hugging Face Transformers

Embeddings & Vector Search

PostgreSQL

Git

Gemini API

Scikit-learn

Vercel

TensorFlow

BART

Groq

Express.js

Node.js

Pandas

Docker

CI/CD

AWS

Vector Search Analysis

Gemini 2.5 Flash

JSON Extraction

Prompt Engineering

Performance Profiling

Independent project creation

ROUGE

Languages

Urdu

English

Projects

Legal Case Summarization - Fine-tuned BART

Developeracademic

Fine-tuned BART on lengthy Supreme Court judgments for abstractive summarization and evaluated the outputs with ROUGE.

PythonHugging Face TransformersBART

View the code

Naano Rebuild - B2B Creator Marketplace

Developerpersonal

Rebuilt a two-sided marketplace end to end in 24 hours with AI coding agents (Claude Code): RLS-secured portals, real-time messaging, Gemini matching, and transactional wallet top-ups.

Next.jsTypeScriptSupabaseGeminiClaude Code

Visit the site View the code

AI Resume Analyzer

Developerpersonal

AI Resume Analyzer built with FastAPI and Llama 3.3 via Groq.

FastAPILlama 3.3Groq

LSTM Stock Forecasting

Developerpersonal

Stock forecasting model using Long Short-Term Memory (LSTM) networks.

TensorFlowLSTM

SmartML

Developerpersonal

No-code AutoML frontend built on AWS.

AWSAutoML

InvoiceFlow - LLM Invoice Extraction

Solo Developerpersonal

Built and deployed solo: Gemini extracts invoice fields from PDFs and photos through a Postgres-backed job queue; a separate validation pass rechecks arithmetic and detects duplicates before human review. Measured extraction accuracy with an evaluation suite on a labeled test set; 30+ automated backend tests.

PythonFastAPINext.jsSupabaseGeminiPostgres

Visit the site View the code

AnswerTrail - Cited AI Support Assistant (RAG)

Solo Developerpersonal

Built solo: a RAG support assistant (FastAPI, pgvector, Gemini) that answers only from help articles, cites the exact passage, and declines when they don't cover the question; private chats enforced in database queries. Measured on 60 hand-written questions: 0 of 20 out-of-scope questions got an invented answer, ∼2.3 s typical response; 126 offline tests plus 14 against Postgres. Designed dev/test/hard eval splits (near-misses, false premises, prompt injection) and showed similarity thresholds alone cannot separate answerable from unanswerable questions; built an atomic document ingestion pipeline.

PythonFastAPINext.jspgvectorGeminiPostgresVercelRender

Visit the site View the code

Education

BS Computer Science

University of Central PunjabComputer ScienceGPA 3.952022 – 2026

Coursework

Deep LearningNatural Language ProcessingArtificial Intelligence

Recognition

Awards

100% Merit Scholarship

University of Central Punjab2022

Full merit scholarship for BS Computer Science

Hire AmbreenReach out about a role, a contract or a conversation.For recruiters

Ambreen's twin is AI, it can make mistakes.

More people on Kanvis

Browse more people Create your own Kanvis