Alation
Data & AI Engineering Intern
Production RAG and multi-agent systems on AWS Bedrock. Retrieval pipelines across Salesforce, Snowflake, Slack, PostgreSQL, and FAISS. Tools supporting 150+ users.
Product-minded AI / ML engineer
I build intelligent systems that hold up in the real world.
Production RAG, agentic workflows, computer vision, and the data infrastructure beneath them—designed around actual people and actual work.
Currently
Building enterprise AI systems at Alation and completing an MIDS at UC Berkeley.
RAG*AGENTS*VISION*DATA SYSTEMS*RAG*AGENTS*VISION*DATA SYSTEMS*
Selected workPublic archive / 2024—2026
(01)
Local Post-Training Lab
Full-parameter fine-tuning, locallyA reproducible lab for full-parameter SFT of code models on Apple Silicon, with executable repair tests that measure whether generated code actually works.
PyTorch / MPS / SFT / Execution evals(02)LIGO Parameter Inference
Collaborative scientific ML researchAn in-progress study of whether neural posterior estimation stays calibrated when gravitational-wave detector noise shifts from LIGO's O3a run to O4a.
PyTorch / Normalizing flows / SBI / LIGO(03)LocalDex
Local-first retrieval infrastructureA private vector indexing toolkit with pluggable embeddings, chunking strategies, and retrieval pipelines for RAG workflows.
Python / Embeddings / Vector search(04)SOC2 Copilot
Evidence-first compliance agentsA read-only AWS evidence collector and secure-code analysis workflow that turns deterministic cloud checks and Semgrep findings into reviewable compliance output.
Python / AWS Bedrock / Strands / Semgrep(05)Watershed Transformation
Geospatial image segmentationA computer-vision pipeline used to identify regional vegetation loss across more than 300 satellite images.
OpenCV / Python / Geospatial MLI work across the ML stack—from training and evaluation to retrieval, infrastructure, and the product people use.
Scope of work / 01ExperienceFrom research to production
Mefferdi
Machine Learning / Cloud Engineer
Architected an AWS data lakehouse across S3 and Glue for evolving telemetry schemas, then built batch simulation pipelines to surface perception-system failure modes before deployment.
uGRIDD
Machine Learning Engineer
Trained YOLO and SAM detection and segmentation systems for road infrastructure. Improved precision and mAP by 22% through dataset balancing, feature engineering, and evaluation against real-world edge cases.
Applied ML Research
Researcher · University of San Francisco
Multimodal and geospatial prototypes, approximately 90% vegetation-loss detection accuracy, and 40% lower segmentation latency.
EducationTechnical foundation
UC Berkeley
Master of Information and Data Science
Graduate study spanning machine learning, data systems, experimentation, and applied AI.
University of San Francisco
Analytics · Computer Science
GPA 3.81. Foundation in statistical modeling, software engineering, and applied machine learning.
Profile
Research instincts. Product judgment. End-to-end ownership.
I’m most useful where the problem is still ambiguous. I talk to users, define the evaluation criteria, build the system, and stay close enough to production to learn where it breaks.
Applied AI
RAG, agents, tool calling, embeddings, model evaluation
ML
TensorFlow, PyTorch, OpenCV, Hugging Face, scikit-learn, computer vision
Systems
FastAPI, PostgreSQL, Snowflake, FAISS, batch and streaming
Cloud
AWS Bedrock, Fargate, S3, Glue, Lambda, SQS, SNS
Languages
Python, TypeScript, SQL, C++, JavaScript
Open to AI / ML engineering roles
Have a hard problem?
Let’s talk about the part that has to work outside the demo.
aaronrassiq@berkeley.edu LinkedIn GitHub