DAMG 7374 Course Labs
Five graduate labs on GenAI in data engineering, authored as TA
The lab material for Northeastern's DAMG 7374, Data Engineering: Impact of Generative AI with LLMs, which I authored and maintained as teaching assistant under Professor Kishore Aaradhya. Five labs take graduate engineers from Snowflake fundamentals through dbt into agent routing, parallel execution and multi-agent reflection, the same patterns I build with.
Verified: each of these is checkable in the repository
- 5
- labs authored and maintainedLab-1 through Lab-5
- 26
- engineers across 9 concurrent team projectsSpring 2026 cohort
- 4
- GitHub stars, my most-starred repositorystudents used it
Teaching agent architecture to people who have just learned dbt is a sequencing problem. Go too abstract and they build things they cannot debug; go too concrete and they copy code without understanding the pattern. Each lab has to leave them able to diagnose their own failures.
Five labs in deliberate order. Snowflake fundamentals across five worksheets. A full dbt project with models, seeds, tests, macros and snapshots, plus setup and teardown scripts so a broken environment is recoverable. A RAG and routing lab with a router dispatching to three specialist tools. A parallelisation lab structured as planner, parallel executor and synthesizer over three analyst agents sharing a base class. Finally a multi-agent reflection pipeline.
Warehouse fundamentals before any agent code
Students who cannot trace a record through transformation layers cannot debug a retrieval pipeline either. Agentic failures look like magic until you can inspect the data underneath.
01Lab 1. Snowflake fundamentals
Five SQL worksheets building from warehouse basics to query profiling.
02Lab 2, dbt
A full project with models, seeds, tests, macros, analyses and snapshots, plus Snowflake setup and cleanup scripts so a broken environment is always recoverable.
03Lab 3. RAG and routing agent
A router dispatching to three specialist tools (retrieval, maps and news) over a Snowflake connection.
04Lab 4. Parallelisation with LLM agents
Planner, parallel executor and synthesizer orchestrating market, purchase and review analysts that share a common base class.
05Lab 5. Multi-agent reflection
A pipeline where agents critique and revise each other's output before a result is returned.
Why it is built this way
01Warehouse before agents
Two labs of Snowflake and dbt before any agent code. Students who cannot trace a record through transformation layers cannot debug a retrieval pipeline either, and agentic failures look like magic until you can inspect the data underneath.
02A shared base class in the parallelisation lab
The three analysts inherit from one base, so the lesson is the orchestration pattern (planner, executor, synthesizer) rather than three variations of prompt writing.
03Teardown scripts alongside setup
Every lab can be reset. Students break environments, and a recoverable environment is the difference between a bad afternoon and a lost week.
04Evaluation rubrics as measurable checks
I wrote rubrics that turned vague quality questions into checks teams could apply to their own systems, which is the same discipline I use on my own projects.
Administering the Snowflake environment behind nine simultaneous team projects taught me more about role-based access and warehouse sizing than my own projects had, because nine teams find every sharp edge at once. And explaining agent orchestration to people who would immediately try to break it forced a precision in my own understanding that building alone had not.
Next project
Gatekeeper
A multi-tenant AI assistant where Postgres, not the model, decides what you see