Imsa Zulfiqar

Data & AI Engineer

I build the data, software and AI systems that make people's work easier.

Python
PySpark
FastAPI
PostgreSQL
ClickHouse
Palantir Foundry
OpenAI
TensorFlow
PyTorch
scikit-learn
pandas
TypeScript
Next.js
React
Kubernetes
Git

About me

Hello, I'm Imsa

I'm a Data & AI Engineer. Some days that means building data pipelines, some days backend services, and some days AI tools, but it always starts with a real problem someone needs solved.

My favourite part of any project is the first conversation. Clients often know something isn't working before they know what they need, so I listen, ask a lot of questions and work out the real problem and the solution behind it. Even early in my career, as the youngest person on very senior teams, I was in the room with stakeholders, gathering requirements and asking the questions that shaped what we built. Then comes the part I love most: seeing people actually use it.

I've enjoyed leading and working with people for as long as I can remember. At 16 I started my own event management business, where I learned how to run a team and look after clients.

I enjoy mentoring junior engineers and helping them grow in confidence. I've also worked with colleagues across the UK, US, India and China, so I'm comfortable collaborating across cultures and time zones.

Ask me

Pick a question below and I'll answer it.

Tech stack

AI & ML

RAGLLM evaluationCustom evaluatorsPrompt engineeringEmbeddingsVector DBsTransformersFine-tuningNLPComputer visionClaude & OpenAITensorFlow

Data

PySparkSQLPostgreSQLClickHousePalantir FoundryData pipelinesFeature engineeringPredictive modellingStatistical analysisAnomaly detectionR

Engineering

PythonFastAPIREST APIsTypeScriptNext.jsKubernetesCI/CDGitSystem designJavaMATLAB

Product & stakeholders

Stakeholder managementRequirements translationSolution engineeringTechnical consultingProduct demosCross-functional collaboration

My work experience

Forward Deployed Engineer
Avirso · Remote (Contract)
Aug 2026 – Oct 2026
- Partnered with 3 client stakeholders, including a Senior VP at a Fortune 10 healthcare company, to build a unified view of $5M+ in year-to-date cloud and AI spend, helping them see which platforms were driving costs and make informed decisions.
- Built ingestion pipelines and a standardised cost data model for billing data from 5+ platforms, including Azure, GCP, Microsoft Copilot, Anthropic Claude and GitHub, mapping each provider's data to FOCUS, the industry standard for cloud cost data, so costs could be compared like for like.
- Contributed to database design and implemented SCD Type 2 logic for discount data, so the history of each platform's discounts was kept and past costs were reported at the right rate. Data was served from ClickHouse, a columnar database that kept analytics fast across very large billing datasets, and deployed through CI/CD on GitHub Actions.
- Worked with a business consultant to define KPIs for 4 user personas, from executive leadership to day-to-day users, so each role saw the metrics that mattered to them.
PythonFastAPIClickHousePostgreSQLKubernetesNext.jsGitHub ActionsFinOpsSCD Type 2
Data & AI Engineer
AbbVie · London
Oct 2024 – Feb 2026
RAG-based market research tool
- Took a multilingual RAG platform on Palantir Foundry from proof of concept to production and led it within a few months, automating analysis across confidential market research reports of 800-1000+ pages, saving the research team hundreds of hours of manual searching while complying with strict data handling requirements and role-based access controls.
- Led the platform's expansion from the EU to the US and China, scaling from 100+ initial users to a global user base across 10 languages, 300+ documents and ~2,000 queries, with separate repositories for each market.
- Worked on the automated ingestion pipeline that processed newly uploaded documents page by page, extracted text, tables and charts with OCR-enabled models, generated page-level metadata and tags, and stored the structured content for retrieval.
- Implemented a tag-driven retrieval workflow where user queries were converted into tags and matched against document content using semantic similarity, with source citations alongside answers so users could verify the underlying evidence.
- Led Phase 2 retrieval improvements using query normalisation, lemmatisation, tag boosting, and prompt guardrails to reduce irrelevant and cross-market results.
- Contributed to developing an LLM as a judge evaluation suite, comparing generated responses against reference answers and assessing relevance, accuracy and groundedness to systematically measure response quality and identify areas for improvement.
AI coaching analytics
- Developed an AI-powered coaching platform that helped managers evaluate coaching sessions with their sales representatives against business-defined frameworks, identify skill gaps, and highlight targeted development areas.
- Built a multi-stage AI pipeline that transcribed recorded coaching sessions, cleaned and enriched transcripts to correct domain-specific terminology, and passed the contextualised output through AI agents for framework-based evaluation and recommended next steps.
- Enriched coaching analysis with speaker identification, sentiment, statement classification, and speaker-balance metrics, turning unstructured conversations into structured insights for managers and sales representatives.
Data engineering
- Standardised development practices across an unstructured PySpark codebase by introducing consistent Git and CI/CD workflows, reducing code duplication by 40%+.
- Designed a reusable feature registry of 25 tested PySpark functions, published as an internal Python library used by 20+ engineers, reducing development time for repeated feature work by about 50% while improving reliability and debugging.
Award for AI Work · Data & Analytics Townhall · Dec 2025Innovation Award · Commercial BTS Recognition Wall · Jul 2025
RAGPalantir FoundryOCRLLM evaluationAI agentsPySparkCI/CD
Machine Learning Engineer
Syslab.ai · Remote
Jul 2021 – Jun 2022
- Developed an adaptive recommendation system that teaches young students by personalising their study paths based on their performance.
- Worked with schools and educators to understand how students were assessed, and used those findings to define how performance and learning gaps should be represented in the platform.
- Curated and structured student performance data alongside a library of learning content, including videos, exercises and interactive activities, to support personalised learning.
- Built the recommendation module using linear regression on quiz performance to identify individual skill gaps and recommend the most relevant content from the learning library.
PythonMachine learningRecommender systemsLinear regression

Professional development

McKinsey Forward Learner

McKinsey & Company

Oct 2026 – Present

Selected for McKinsey Forward, a programme focused on problem-solving, communication, adaptability and leadership.

What people say

Recommendations from people I've worked with, quoted from LinkedIn.

“She is undoubtedly one of the most exceptional AI professionals I have had the opportunity to work with.”

Piyush RaiSoftware Engineer specialising in AI and Machine LearningWorked with Imsa for over a year on AI & ML projects

“She pays close attention to detail, thinks problems through and asks the right questions.”

Manal Rizwan QureshiData EngineerWorked with Imsa on the same team

See all recommendations on LinkedIn

Selected projects

Production-style AI tools and applied machine learning I've built outside work.

citedrag · Document Q&A

Upload documents, ask questions, and get answers that cite the exact passage behind every claim. Market-aware BM25 retrieval, Claude's citations API, an evaluation suite for search quality, and a web app with private, in-browser uploads.

Live demo GitHub

VoiceQuery LLM Pipeline

An end-to-end voice assistant pipeline. It records a spoken query, transcribes it with Faster-Whisper, generates a response with a Hugging Face LLM, and speaks the answer back with Google TTS.

GitHub

ASL Sign Translator

Recognises American Sign Language letters and common words, then reads the translation aloud. CNN with residual blocks, depth-wise separable convolutions, spatial and transformer-style attention, plus mixup augmentation. Reached 80% accuracy on combined ASL colour and MNIST fingerspelling data.

GitHub

See everything on GitHub