SOKHENG(1)User CommandsSOKHENG(1)
NAME
sokheng — applied scientist, machine learning & optimisation
SYNOPSIS
sokheng [--research] [--engineering] [--teach] problem ...
DESCRIPTION
Builds machine-learning systems, from LLM pipelines to models that know how unsure they are, and writes down the mathematics that makes them work.
Currently studying for an M2 in Data Science at Université Paris-Saclay while working as an AI & Data Science apprentice at IFP Energies nouvelles.
Trained as an engineer in Cambodia, now based in Paris.
HISTORY
$ git log --graph --all --date-order
859686f2026-09(HEAD -> work)merge study: alternance, AI & Data Science apprentice at IFP Energies nouvelles
9596a022026-06Data Science intern, GI2: investor matching for Pitchtree
e1112962026(study)M2 Data Science (health, insurance, finance), Université Paris-Saclay
d1111032025mv ~ /paris · M1 Applied Mathematics, ENSIIE · conformal prediction research with N. Brunel
a596b952024-05Software Engineer, CDC Cambodia: CI/CD, face verification, MCP + LLM
c110f702024B.Sc. Electronics & Automation Engineering, ITC
3b4ca642019initial commit: Institute of Technology of Cambodia
EXPERIENCE
2026-09 → now● running
AI & Data Science ApprenticeIFP Energies nouvelles │Rueil-Malmaison- Builds LLM, RAG and knowledge-graph pipelines that turn decades of experimental archives into something you can search and question.
- Designs an agentic workflow for scientific monitoring: agents that read new work and surface what matters.
agentsknowledge-graphsllmrag
2026-06 → 2026-09✓ exited 0
Data Science InternGI2 │Paris- Built the investor-matching and persona pipeline for the Pitchtree platform: which investors fit which startup, and why.
llmmatchingpython
2024-05 → 2025-08✓ exited 0
Software EngineerCDC Cambodia │Phnom Penh- Built and ran CI/CD pipelines for production services.
- Shipped face verification into production.
- Built MCP + LLM systems that connect language models to internal tools.
ci/cdcomputer-visionllmmcp
EDUCATION
2026 → now● running
M2 Data Science · Health, Insurance, FinanceUniversité Paris-Saclay │apprenticeship track, finance option- Taken alongside the IFPEN apprenticeship: study and work, merged.
financemachine-learningstatistics
2025 → 2026✓ exited 0
M1 Applied MathematicsENSIIE │Évry- Research project on weighted conformal prediction under prompt shift, supervised by Prof. Nicolas Brunel.
conformal-predictionoptimisationprobability
2019 → 2024✓ exited 0
B.Sc. Electronics & Automation EngineeringInstitute of Technology of Cambodia │Phnom Penh- Control, embedded systems and signal processing.
controlembeddedsignal-processing
OPTIONS
--llm, --ragLLM systems, retrieval-augmented generation, knowledge graphs, agentic workflows, MCP.
--mlopsFastAPI services, MLflow experiment tracking, CI/CD pipelines.
--visionFace verification, license-plate recognition.
--uncertaintyConformal prediction, density-ratio estimation.
--mathOptimal control, stochastic calculus, deep learning theory.
--mobileFlutter (Sprout).
--hardwareElectronics, automation and embedded systems background.
EXAMPLES
$ sokheng --rag ~/archives/1998/*.pdf
Turns decades-old lab reports into a knowledge graph you can ask questions.$ sokheng --uncertainty --alpha 0.1 model.pt
Wraps a model’s predictions in intervals that hold 90% of the time, even when the prompts drift.$ sokheng --teach backprop
Writes a chapter: the derivation by hand first, then the NumPy code.$ sokheng --mlops ./trade-ml-api
Ships a model as a FastAPI service, with MLflow tracking and CI/CD.EXIT STATUS
0Problem solved.
1Needs more data.
2Interesting enough to become a blog post.
130Interrupted by the cat.
LANGUAGES
Khmer (native) · [OTHER LANGUAGES AND LEVELS]
FILES
~/writingcourses and write-ups
~/projectsthings built and shipped
publications.bibreports and challenge entries
SEE ALSO
github(1) · linkedin(1) · email(1) · writing(7)
BUGS
Occasionally overfits to interesting problems. Feeds the cat before replying to email.
portfolioSOKHENG(1)