Aaron Lim

AI/ML • Middle • Fremont, USA • UTC-08:00 • Relocation • Work permits: United States • English: C2 • Available now • from $90k per year

Software Engineering Intern

Python

JavaScript

SQL

TypeScript

Node JS

Kotlin

Experimentation & Evaluation: 4/10

Invite to interview

Download CVCV

Overview

Technical skills

Projects

Timeline

Roles

Overview

LLM engineer (senior-level) specializing in retrieval-augmented meal-planning systems with safety-focused evaluation. The strongest proven skill is building evaluation and RAG pipelines as shown by src/evaluator.py and the retrieval + filtering logic in src/retrieval.py. There is little to no public evidence of custom model training, advanced MLOps (deployment/monitoring), or efficiency optimization work.

Technical skills

Languages

Python

JavaScript

SQL

TypeScript

Node JS

Kotlin

AI/ML

Streamlit

LLM

LangChain

Transformers

Other

Requests

Azure

Nest.JS

MySQL

Exposed

Ktor

HikariCP

Pytest

JUnit

Ollama

Docker

Git

Rest API

Sentry

Embeddings

Docker Compose

Swagger

RAG

Projects

Food Planner

Jul 2026 to Present 2 Months

Python

LLM

Ollama

Sentence-Transformers

Pytest

Docker Compose

Docker

Chroma

Tournament Pilot

Jun 2026 to Present 3 Months

Kotlin

Ktor

HikariCP

Rest API

MySQL

JUnit

Timeline

University of California (Davis Campus)

Bachelor's Degree • Computer Science

2023–2026 Davis, California

Software Engineering Intern • Junior

Flip-ers • Internship

Seoul In office

Debugged TypeScript issues in a NestJS backend and validated REST API behavior using Swagger. Integrated Sentry for real-time production error tracking to speed up incident response. Troubleshot Azure App Service configuration problems, including database connectivity checks, to restore reliable application behavior. Wrote onboarding documentation and bug reports in GitHub Issues to improve team ramp-up.

TypeScript

Swagger

Rest API

Azure

Sentry

Git

Middle AI/ML Engineer Confidence: Medium LLM Engineer

Model Architecture & Training

1/10

How well models are designed and trained

Minimal model-building or training work; system relies on hosted/packaged models and embedding functions rather than custom architectures or training loops.

Evidence

gigaaaron/food-planner/src/evaluator.py: uses call_ollama() to call an LLM for evaluation gigaaaron/food-planner/src/retrieval.py: uses SentenceTransformer embedding function (MODEL_NAME = 'all-MiniLM-L6-v2')

Data Pipeline & Feature Engineering

3/10

How data is prepared for models

Clear, purposeful data retrieval and filtering pipeline with query construction, over-fetch + allergen filtering and tests ensuring metadata fields; not a large-scale ETL system but solid feature engineering for RAG.

Evidence

gigaaaron/food-planner/src/retrieval.py: build_query, retrieve_recipes, filter_by_allergens, get_recipes_for_user gigaaaron/food-planner/tests/test_retrieval.py: many unit tests validating query contents, allergen filtering and metadata fields

Experimentation & Evaluation

4/10

How results are measured and tested

A structured evaluation pipeline with multiple scoring functions, per-category summaries, and automated result saving; good test coverage around desired behaviours and safety checks.

Evidence

gigaaaron/food-planner/src/evaluator.py: run_eval, score_keyword_hit, score_allergen_safety, score_context_precision, score_answer_relevance gigaaaron/food-planner/tests/test_meal_generation.py: unit tests that validate output structure, safety and goal alignment

MLOps & Deployment

2/10

How models are shipped to production

Basic serving and UX via Streamlit and local feedback logging; persistent ChromaDB usage but limited MLOps concerns such as deployment automation, monitoring, CI/CD or model/version management.

Evidence

gigaaaron/food-planner/streamlit_app.py: Streamlit app entry and wiring to UI components gigaaaron/food-planner/src/retrieval.py: chromadb.PersistentClient usage (CHROMA_DIR)

Computational Efficiency

1/10

How efficiently computing resources are used

No evidence of GPU/memory optimizations, quantization, batching strategies or profiling; standard library and packaged embeddings are used without efficiency engineering artifacts.

Evidence

gigaaaron/food-planner/src/retrieval.py: uses off-the-shelf SentenceTransformer embedding function without custom performance code gigaaaron/food-planner/src/evaluator.py: simple loop with time.sleep, no batching or async inference

Research Depth & Innovation

1/10

Depth of research and new ideas

Limited research depth or novel algorithmic work; the project applies standard RAG patterns, safety checks and evaluation heuristics but contains no custom layers, paper implementations or ablation studies.

Evidence

gigaaaron/food-planner/src/evaluator.py: pragmatic scoring heuristics rather than experimental ablation gigaaaron/food-planner/src/retrieval.py: standard Chroma + SentenceTransformer usage

Verified artifacts

evuniuzhi/studio-ghibli-csp• PR

Expertise

RAG• Middle

Industries

Artificial Intelligence• Middle

Food & Beverages• Middle

Technologies

Python• since 2025 • Middle

LangChain

Embeddings

Ollama

Transformers

LLM

RAG

Streamlit

Requests

Flask• mentioned only

TensorFlow• mentioned only

Recommendations

Develop retrieval-augmented generation (RAG) systems and safety-aware prompt/eval pipelines for consumer-facing assistants.
Implement and extend evaluation dashboards and automated scoring systems (using the existing evaluator.py as the base).
Build Streamlit prototypes and user-feedback loops for rapidly validating LLM-driven product features in the food/nutrition domain.
Hardening allergen and safety filters and integrating vector store versioning or simple model/versioning for controlled deployments.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

food-planner music-genre-ml