April 2024 Summaries

Cron jobs are a way to automate work by running scripts at regular intervals, such as every hour or daily. They use a specific syntax to specify the schedule, which includes five numbers representing minute, hour, day of month, month, and day of week. Cron uses the host machine's time zone, typically UTC, and can be used for various tasks like data jobs, sending reports or alerts, and system maintenance. However, cron has limitations, including being primarily made for shell scripts, lacking monitoring, and requiring cloud provisioning. Modal is an alternative to cron jobs that allows scheduling Python code directly with basic monitoring capabilities.

With just a handful of examples, fine-tuned open-source embedding models can provide greater accuracy at a lower price than proprietary models. Custom models matter for companies like Netflix and Spotify that use data to improve their recommendation systems. Large pre-trained models with permissive licenses have simplified the bootstrap step, allowing organizations to start with these models and expect them to perform reasonably well on their task. Fine-tuning kicks off the data flywheel by accumulating data quickly, which can lead to better performance than the off-the-shelf model. The process of fine-tuning involves design decisions such as finding or creating a dataset, choosing a base model, and acquiring training infrastructure. Running a grid search over fine-tuning hyperparameters is an effective way to explore experimental parameters, and Modal's autoscaling infrastructure can be used to scale experiments in parallel. Even with just a few hundred examples, it's possible to beat proprietary models on a simple question-answering task, and moving forward, the next step would be to operationalize this process to collect more data and iterate on the model.

ETL moves data from source systems into analytical warehouses, typically using managed connectors such as Fivetran or orchestration tools like Airflow, while the post presents Modal as a usage-priced, code-focused alternative for large transfers and custom scheduled jobs. In one example, a Modal function extracts ClickHouse usage metrics and batch-loads them into Snowflake across five parallel date-based jobs, reportedly transferring 12 million rows in 16 minutes for about $0.29 in compute and memory costs; the comparison estimates that a comparable Fivetran sync would cost roughly $3,300, though it notes that custom code also requires engineering time. A second example retrieves GitHub usernames from Snowflake and queries the GitHub API to enrich user records with listed company information, with Modal Secrets used for credentials and a cron schedule available for daily execution. The post argues that custom code can improve flexibility, reduce vendor lock-in, and lower costs, while acknowledging that conventional ETL services remain useful for common low-to-medium-volume connectors and that Airflow-like orchestration platforms are better suited to long-running, business-critical, multi-stage pipelines requiring retries, caching, and detailed observability.

Modal, a serverless platform serving AI and data teams, was named third among 10 early-stage companies on the sixth annual Enterprise Tech 30 list, which identifies promising private enterprise technology companies. The selection process considered nearly 600 venture-backed firms and involved 112 venture capitalists and 36 corporate development teams, with companies grouped by total capital raised. Modal CEO Erik Bernhardsson credited the recognition to the company’s team, customers including Suno and Ramp, and investors, while emphasizing plans to expand its platform. Enterprise Tech 30 founder Peter Wagner said the ballot-based list recognizes companies viewed as likely to drive business transformation and potentially achieve major exits or public offerings.