Turn a raw media archive on Backblaze B2 into a versioned segmentation dataset. Drop click/box prompts, run Meta's Segment Anything 2 (SAM 2) locally to mask images and propagate across video, and write COCO RLE, mask PNGs, and cut-outs back to B2 — no second API key.

Sample app that splits audio into vocals, drums, bass & other stems with Demucs (htdemucs) locally and archives every track + stem to Backblaze B2 over the S3-compatible API. Next.js 16 + FastAPI, browse & stream stems from the bucket.

Archive, search, and play back Frigate NVR surveillance footage on Backblaze B2. Ships Frigate's recordings, event clips, snapshots, and a searchable object-detection index to B2 (S3-compatible), then searches the timeline and streams clips via presigned URLs. No database. Sample app.

Batch object-detection pipeline that runs Ultralytics YOLO11 (detect/segment) over an image/video library on Backblaze B2 S3-compatible storage, writing COCO JSON, annotated frames, and per-class instance crops back to B2.

Deep-research AI agent that turns a question into a long-form cited report and caches every source page, screenshot, and report to Backblaze B2 (S3-compatible object storage) as the agent's durable memory. Claude Sonnet 4.6 tool-use loop, Next.js + FastAPI. A Backblaze B2 sample app.

Run SQL directly against Parquet, CSV, and JSON files in Backblaze B2 with DuckDB's httpfs extension — query-in-place with no ETL or warehouse, and materialize results back to B2 as Parquet. Next.js + FastAPI sample app.

Backblaze B2 Partner Portal — example portal for the Partner API

Multimodal semantic search over images and PDFs with no vector-database server: the LanceDB vector store and ANN index live directly on Backblaze B2. Local CLIP embeddings, S3-only.

Computer-vision sports-highlights pipeline: detect, track, clip, annotate, and AI-summarize the action in a sports video, with Backblaze B2 (S3-compatible) as the sole data store. Built with Roboflow Inference + Supervision, ffmpeg, Claude, FastAPI, and Next.js.

Apache Iceberg feature store and data lakehouse on Backblaze B2 — versioned feature tables, schema evolution, snapshot rollback, and DuckDB time-travel queries via PyIceberg, with B2 as the warehouse (no data warehouse to run)