Skip to content
MI

Currently: AI research at the Machine Intelligence Lab

Building AI products that solve real human problems.

I’m Maryam, a software engineer in training and an AI research assistant. I build systems end to end, from the data and the model calls to the evaluation that shows whether they actually work, and I design them around the person on the other side of the screen.

Explore my workLinkedIn(opens in a new tab)GitHub(opens in a new tab)
Portrait of Maryam Ibaaichou wearing a black hijab and blazer against a deep burgundy background
Maryam Ibaaichou

01Experience

Proof of work, not a list of skills.

An industry internship in big data, and ongoing AI research at my university lab. Both start with a real problem and end with what I learned.
01Big Data Engineering Internship
Role
Big Data Engineering Intern
Company
Suncaper (Chengdu Suncaper Data Co., Ltd.)
When
2026 · team project

SkyQuery: asking flight data in plain language

At Suncaper, a big-data company in Chengdu, I worked on a flight meta-search and analysis platform built on 15M+ Expedia itineraries, with a conversational interface that turns a traveller's question into a database query and shows the answer as a table, chart or route map.

Problem
Flight prices feel arbitrary. Travellers watch fares jump, can't tell whether to buy now or wait, and sometimes chase low prices that can't actually be booked.
Thinking
Start from the questions people really ask: Is this fair? Should I wait? Why is this route so expensive? Answer them with data, and let people ask in their own words instead of in SQL.
What I built
Analyses of pricing, hub premiums, buy-or-wait booking windows, and ghost fares: sudden price swings that pressure people into panic-buying. I also built the conversational front end: a chat UI that sends questions to the text-to-query backend, displays the generated HiveQL, and automatically renders results as a table, chart or route map.
Learning
A natural-language interface only earns trust if it shows its work. Every answer displays the query it ran, so a person can check it instead of taking it on faith.

Technology

  • Hadoop
  • Hive
  • PySpark
  • JavaScript
  • Leaflet
  • REST APIs
  • Agile / JIRA
View on GitHub(opens in a new tab)
Which destinations from JFK are cheapest on average?

Generated HiveQL

SELECT destinationAirport,
  AVG(totalFare) AS avg_fare
FROM itineraries
WHERE startingAirport = 'JFK'
GROUP BY destinationAirport
ORDER BY avg_fare LIMIT 5;
TableChartRoute mapchosen automatically from the result columns
Interface flow · illustrative query
02AI Research
Role
Research Assistant
Lab
Machine Intelligence Lab, Sichuan University
When
Aug 2026 — present

Research on reliable multi-agent AI

At Sichuan University's Machine Intelligence Lab, I work on multi-agent AI systems for medical-research tasks, with a focus on evaluation: making it possible to see whether AI agents actually got it right, not just whether they answered.

Focus

  • Multi-agent systems
  • Medical AI
  • AI evaluation
  • Reliability
Ongoing · details to be shared once published

Closed loop

Did the agent
get it right?

PlanActEvaluateLearn
The idea, not the results · research in progress

02About me

Engineer by training. Product-minded by instinct.

I got into software engineering because I liked making things work. I’ve stayed because the harder question is who they work for.

I’m an international student at Sichuan University, studying Software Engineering on a full merit scholarship and graduating in 2027. My foundation is the core of the discipline: data structures, algorithms, databases, operating systems and software architecture.

In 2026 I joined Suncaper, a big-data company in Chengdu, as a Big Data Engineering Intern, working on flight-data analysis and a conversational search interface. The same year I became a Research Assistant at the university’s Machine Intelligence Lab, working on multi-agent AI. Between the two, I learned to ask one question of every AI feature: how would we know if it’s wrong?

I speak six languages: Arabic and Amazigh natively, French and English fluently, Chinese and some Turkish. Living and studying across cultures keeps me asking who a product is really for, and who it quietly leaves out.

Where I’m going: work where engineering and product meet. That means building AI products, defining what “good” means for them, and measuring whether they get there.

languages spoken
6
merit scholarship
Full
graduating
2027

Languages · 6

  • ArabicNative
  • AmazighNative
  • FrenchFluent
  • EnglishFluent
  • ChineseHSK 4
  • TurkishA2

My journey

  1. 2023

    Started Software Engineering at Sichuan University

    Full Merit Scholarship

  2. 2024

    Belt and Road Culture & Health summer programme

    Southwestern University of Finance and Economics, Chengdu

  3. 2024

    Digital Journalism Workshop, Turkey

    International programme · 80% merit-funded

  4. 2026

    Big Data Engineering Intern at Suncaper

    Flight meta-search & conversational analytics

  5. 2026

    Research Assistant, Machine Intelligence Lab

    Multi-agent AI & evaluation

  6. 2027

    Graduating, B.Eng. Software Engineering

    Heading toward AI product & applied AI roles

03Product thinking

How I approach a problem, before I write code.

I don't have years of product management behind me. What I do have is a habit, formed by building and evaluating real systems, of asking the product questions first. Here's how I work through a problem.
  1. 01

    Understand the user

    Who is on the other side, and what are they actually trying to decide?

    How I do itStart from the real question. In SkyQuery, travellers didn't want fare tables. They wanted to know whether to buy now or wait, which became a booking-window heatmap.

  2. 02

    Define the problem

    Write down what success looks like before writing code.

    How I do itOne sentence, one outcome you can measure. If a feature can't be tied back to it, it waits.

  3. 03

    Explore constraints

    Constraints are design input, not obstacles.

    How I do itData access, privacy, speed and cost all shape the design. In SkyQuery, queries ran over 15M+ rows, so the interface had to make waiting feel clear, not broken.

  4. 04

    Build

    Get the smallest complete version working first.

    How I do itA rough end-to-end path that runs beats a polished piece that doesn't connect to anything. Polish comes after it works.

  5. 05

    Evaluate

    Separate “it responded” from “it was right”.

    How I do itEspecially with AI, a complete-looking answer can still be wrong. I measure both, and compare against a simple baseline.

  6. 06

    Iterate

    Let failures choose the next step.

    How I do itThe weakest result shows where to look next, so that's where the next round of work goes.

The loop doesn’t end at step six. Evaluation feeds back into understanding the user, which is why I think evaluation is the most underrated product skill in AI.

04Technical skills

The toolkit behind the work.

Engineering
  • Python
  • Java
  • JavaScript
  • SQL
  • REST APIs
  • FastAPI
  • Spring Boot
  • Git & testing
AI & evaluation
  • Multi-agent systems
  • LLM APIs
  • Evaluation design
  • Neural networks
  • scikit-learn
Data
  • Hadoop & Hive
  • PySpark
  • Analysis & visualisation
Product
  • Requirements framing
  • Agile / JIRA
  • Data storytelling

05Contact

Building something that should work for real people?

I’d like to hear about it. I’m open to conversations about AI products, evaluation, research and internships.