Skip to content

Laiba Khawar

AI Engineer

AI Engineer

Generative AI, RAG, graph memory, and agentic workflows, from model pipelines to production services.

View resume

Select a stage to trace it through the network.

Selected work

Systems I designed and built, from agent orchestration and retrieval to models trained from scratch. Each one opens into a case study with its architecture, decisions, and recorded results.
Agentic AIRAG and Knowledge

Veriflow

An agent platform that has to prove every claim it makes before a human signs off.

FastAPIPydantic v2SQLAlchemy 2PostgreSQL 16pgvectorRedis
EDIConnect mapping interface
Generative AIData Engineering

EDIConnect

Hybrid rule-based and LLM-assisted mapping for B2B EDI documents.

PythonLLaMA 3.1Sentence embeddingsFAISS
RAG and KnowledgeGenerative AI

Netflix Q&A Assistant

A RAG chatbot that knows when to look something up exactly instead of searching for it.

LangChainChromaSentence Transformers (all-MiniLM-L6-v2)Groq
NLPGenerative AI

English to Urdu Translation

A Transformer trained from scratch on a 24,525-pair parallel corpus.

PyTorchnn.TransformerSentencePieceHugging Face tokenizers
Generative AIComputer Vision

Video Understanding Pipeline

Speech, frames, and a language model, run in parallel to summarise a video.

Whisper smallViT-GPT2 captioningPhi-3 mini 128kHugging Face Transformers
Confusion matrix of the fine-tuned ViT on CIFAR-10
Computer VisionMachine Learning

CNN vs Vision Transformer

What transfer learning buys on CIFAR-10: 69% from scratch, 97% fine-tuned.

PyTorchtorchvisiontimmViT-Tiny

Inside the systems

The same projects, opened up. Pick one to trace how data moves between its components.

An agent platform that has to prove every claim it makes before a human signs off.

Read the case study
How it fits together

Select a component to see what it does. Blue packets show the direction data moves.

  • Objective to Planner
  • Planner to Data agent (tasks)
  • Planner to Retrieval
  • Planner to Risk agent
  • Data agent to Investigator
  • Retrieval to Investigator (evidence)
  • Risk agent to Investigator
  • Investigator to Verifier (claims)
  • Verifier to Report (supported)
  • Report to Human approval (actions)

Run the algorithms

Faithful in-browser reimplementations of notebook projects, running on the original datasets where the repository includes them.
The Mona Lisa, one of the three test images used for segmentation Live demo
Computer VisionMachine Learning

Lazy Snapping Segmentation

Paint a few strokes and let colour clusters separate foreground from background.

NumPyscikit-learn KMeansPillowTypeScript (demo)
Face KNN demo: a test face beside its seven nearest training faces and their distances Live demo
Computer VisionMachine Learning

Face Recognition with KNN

A from-scratch nearest-neighbour classifier on 32 x 32 faces, compared with SVM and Naive Bayes.

NumPyscikit-learnPCATypeScript (demo)
Perceptron decision boundary over two classes Live demo
Machine Learning

Rosenblatt's Perceptron

The original learning rule, implemented from scratch and animated.

NumPyMatplotlibTypeScript (demo)
Convolution versus correlation with an asymmetric kernel Live demo
Computer Vision

Convolution from Scratch

The operation behind CNNs, written by hand and compared with SciPy.

NumPyOpenCVSciPyTypeScript (demo)
Cryptarithmetic demo: column-wise search assigning digits to SEND + MORE = MONEY Live demo
Machine Learning

Cryptarithmetic Search

SEND + MORE = MONEY, solved by search, with the frontier on screen.

PythonA*DFSBFS
Bakery lock simulator: three threads, their choosing flags and ticket numbers Live demo
Systems

Bakery Lock Simulator

Lamport's mutual exclusion algorithm, one memory operation at a time.

C++pthreadsstd::atomicTypeScript (demo)
Bridge simulator: vehicles queued at a single-lane bridge with condition-variable waiters Live demo
Systems

Single-Lane Bridge

Cars, buses, a mutex, and two condition variables.

C++pthreadsCondition variablesTypeScript (demo)
Bitcoin forecasting demo: daily closing price with a chronological train and test split Live demo
Machine Learning

Bitcoin Price Forecasting

Regression baselines, stationarity tests, and walk-forward evaluation on daily BTC prices.

pandasscikit-learnstatsmodelsTypeScript (demo)
Urdu segmentation demo: an Urdu passage split into sentences with boundary triggers Live demo
NLP

Urdu Sentence Segmentation

Rule-based sentence boundaries for Urdu, where punctuation alone is not enough.

PythonregexurduhackSentencePiece
All 41 public projects Open the retrieval lab

Experience

Production LLM and NLP platforms, graph-based customer memory, and the teaching and internship work before that.
  1. Feb 2026 to Present

    AI Engineer, SysReforms International

    Building and optimizing PVS-LLM, a production AI/NLP platform that converts unstructured veterinary and public-health reports into structured, validated insights.

    • Production LLM pipelines for summarization, question answering, concept extraction, semantic mapping, and learning-needs analysis.
    • Automated model-evaluation and QA frameworks combining semantic similarity, NLI, BERT-based metrics, and LLM-as-a-judge scoring.
    • Throughput work through batching, parallelism, model preloading, and resumable checkpointed workflows.
    • Multilingual FastAPI and Streamlit applications for conversational AI, evaluation, reporting, and human review.
    MistralOllamaAzure AISentence TransformersFastAPIStreamlitPostgreSQLSQL ServerDocker
  2. Jan 2025 to Jan 2026

    AI Backend Engineer, Dataspecc

    Built Hyperengage's Customer Memory Graph, an AI infrastructure layer modelling accounts, contacts, commitments, risks, deals, and time-bounded customer relationships.

    • Hybrid retrieval combining pgvector semantic search with graph traversal.
    • LLM-powered services that generated and maintained contextual memory summaries for customer accounts.
    • Data pipelines integrating HubSpot, Salesforce, Segment, Intercom, and Zendesk into a unified graph model.
    • Backend functionality for the Signals Engine and Copilot Q&A: real-time pattern detection and natural-language querying.
    Neo4jMemgraphpgvectorPostgreSQLPythonLLMs
  3. Feb 2025 to Jun 2025

    Teaching Assistant, NLP and Database Systems, FAST-NUCES

    Supported lectures, lab sessions, and assessment for undergraduate NLP and Database Systems courses.

    NLPSQLDatabase Systems
  4. Jun 2024 to Aug 2024

    Machine Learning Intern, Devsort

    Contributed to development, testing, and deployment of ML solutions across data collection, cleaning, preprocessing, exploratory analysis, model development, and evaluation.

    • Used Python, Pandas, NumPy, Scikit-learn, and Matplotlib to surface patterns and data-quality issues, and reported results through visualizations.
    PythonPandasNumPyScikit-learnMatplotlibGit

How I work

AI Engineer building and deploying production LLM and NLP systems, including retrieval-augmented generation, automated model evaluation, graph-based memory, and semantic search. I take models from integration and fine-tuning through to scalable Python services, data pipelines, and cloud deployment.

Across these roles the common thread is the system around the model: retrieval scoped to the right customer or document, evaluation that goes beyond exact matching (NLI, semantic similarity, LLM judges), and long-running pipelines made resumable with checkpoints.

LLM systems

Pipelines, evaluation, and the parts that make model output dependable.

Retrieval and memory

Getting the right context to the model, from vectors and from graphs.

Models and frameworks

The modelling toolkit used across work and projects.

Backend and data

Services and pipelines that carry models into production.

Databases

Relational, document, vector, and graph stores.

Cloud and delivery

Where the services run.

Education

  • BS Data Science
    FAST-NUCES, Islamabad, 2021 to 2025
    Dean's List: Spring 2023, Fall 2024
  • A Levels (Engineering)
    Beaconhouse Margalla Campus, Islamabad

Certifications

  • LangChain: Chat with Your Data
  • AWS Web Application Builder
    AWS
  • AWS Data Pipeline Builder
    AWS
  • Understanding Data Visualization
    DataCamp
  • Intro to Data Science in Python
    DataCamp

Get in touch

Questions about a project, a role, or something to build together. Email is the fastest way to reach me.
Based in
Islamabad, Pakistan

Or email laibakhawar7@gmail.com