Translation Model Benchmark for Multilingual Video Transcripts
A multilingual benchmark comparing Google Translate, DeepL, and Llama Maverick 4 on noisy video transcript data across 15 languages.
nikolasbakalis.com
Nikolas Bakalis
Personal work, research, and technical writing
A focused archive for backend systems work, data infrastructure benchmarks, ML/NLP research, and technical writing that is useful as context for collaborations, applications, and deeper engineering conversations.
Selected research
Reports and artifacts
RDS vs StarRocks 20M Serving and Aggregation Benchmark
A 20 million row benchmark comparing RDS/Postgres serving tables with StarRocks OLAP tables and async materialized views.
Materialized View API Serving Benchmark
An API-level comparison of denormalized RDS tables and StarRocks async materialized views across 100k, 1m, and 10m row scales.
IAB 3.0 Content Classifier Training Report
A training report for an in-house hierarchical IAB 3.0 content classifier built to replace external classification dependencies.
API Framework Benchmark for Data-Intensive Services
A benchmark of API frameworks and datastore access patterns for data-intensive services spanning PostgreSQL, StarRocks, and OpenSearch.
IAB 3.0 Classification and Language Detection Pipeline Proposal
A project proposal for replacing external classification APIs with in-house IAB 3.0 classification and language detection pipelines at media-corpus scale.
Tools
Selected technologies
The stack changes by project, but the recurring work is around reliable data paths, measurable model behavior, and production API surfaces.
- AWS
- PostgreSQL
- StarRocks
- OpenSearch
- Iceberg
- Python
- TypeScript
- Go
- Bun
- Docker
- FastAPI
- Fastify
- Celery
- NLP
- LLM evaluation
Research
Dated work that can be cited
Translation Model Benchmark for Multilingual Video Transcripts
A multilingual benchmark comparing Google Translate, DeepL, and Llama Maverick 4 on noisy video transcript data across 15 languages.
Report index
- Backend performance benchmarks
- Database serving architecture
- ML and NLP classification systems
- Translation model evaluation
New York, NY
About this site
I work on data-intensive backend systems, evaluation workflows, and ML/NLP pipelines where correctness, performance, and operational tradeoffs matter. This site collects dated reports and implementation notes so the work can be reviewed directly instead of reduced to a resume bullet.
