Diakhere Gueye
Applied AI Solutions and UX Research
French citizen | diakhere7@gmail.com
diakheregueye.com | linkedin.com/in/diakhere-gueye | github.com/GueyeD
Profile
I design, build and evaluate AI solutions, from LLM applications to machine learning models, and help organisations adopt them. I work alongside leadership and engineers, and ground every solution in mixed methods research.
Professional experience
Visentis GmbH, computer vision AI startup
AI Product Development Associate
Zug, Switzerland | Jan 2026 - Jul 2026
- Secured client approval for a supervised maintenance pilot of Argos, an industrial AI monitoring system, by building the Python simulator and REST API that tested incident assignment, handover and resolution.
- Made the Argos API hold up when messages arrive late, twice or out of order, so an old reading cannot overwrite a newer one and a retry never repeats an action.
- Found in tests that a normal reading made open incidents look closed, so I required operator confirmation to close.
- Set who owns each Argos incident and its next action at shift change, after interviewing 8 staff across 2 sites, from operators to supervisors, and reviewing their handover records.
- Showed from 12 interviews and 8 observed matches that players could link a call from Orus, an AI officiating system for padel, to a different shot than the system did. Engineers built the replay I specified.
- Wrote the Orus recording and annotation protocols that kept development and evaluation footage apart.
Groupe Afrique Challenge, consulting and executive education firm
Associate Consultant, Research and Strategy
Casablanca, Morocco | Feb 2024 - Dec 2025
- Built an internal LLM reporting tool in Python (FastAPI, PostgreSQL, OpenAI API) that drafts client reports from consultants' notes, with the engagement lead approving each report before export.
- After an early draft reported a proposed date as agreed, split extraction from drafting so every statement keeps its source and is marked as an observation, a proposal, a decision or an action.
- Compared 3 model sizes on 18 development cases for factual support, latency and cost. The smallest one dropped conditions and the largest cost more without better content, so the mid-size model shipped.
- Wrote regression tests for conflicting records, unsupported claims, prompt injection and cross-client access.
- Cut the median time to prepare, draft and check a report from 45 to 28 minutes (38% less) in a controlled test with 6 consultants. In blind review, 11 of 12 tool and 11 of 12 manual reports needed no substantive correction.
- After a 6-week pilot the firm kept the tool, and 4 of the 6 consultants used it for at least 3 of their 4 weekly reports.
- Co-delivered AI training for 100+ executives and managers from 5 African countries.
- Secured board support at a financial institution to cut median funding decision time from 5 to 3 working days.
International Atomic Energy Agency
Machine Learning Intern
Vienna, Austria | Oct 2022 - Jun 2023
- Built TensorFlow and XGBoost baselines to predict plasma disruptions, split by experiment to prevent leakage.
- Co-organised the ITU and IAEA Multi-Machine Disruption Prediction Challenge (460+ registrations) and revised its starter examples and validation guidance after watching users misread the targets and data splits.
Education
University of Edinburgh, MSc Cognitive Science with Merit
2020 - 2021
École Polytechnique, BSc Mathematics and Physics
2017 - 2020
Skills
AI: LLM APIs (OpenAI, Groq), RAG, AI agents, LangChain, FAISS, prompt engineering, structured outputs
Engineering: Python (pandas, FastAPI, Streamlit), TensorFlow, XGBoost, SQL (PostgreSQL), REST APIs, Git
Evaluation: Test datasets, regression testing, human review, error analysis, latency and cost analysis
Delivery: Technical discovery, requirements gathering, proof-of-concept scoping, pilots, stakeholder workshops, human-in-the-loop design, user training, executive presentations, adoption measurement
User research: Interviews, contextual inquiry, survey design, usability testing, experimental design
Languages: French (native), English (fluent), Spanish (intermediate), Arabic (intermediate)