FIELD NOTE / Media Infra
The hidden 6 MB Lambda trap: compression after the limit is too late
Why API Gateway compression can make large responses look safe in testing, then still fail when Lambda enforces the raw payload limit first.
Read the noteEngineering notes from building VideoDB systems and AI agents.
Notes on video infrastructure, retrieval, streaming, editing, and agents from the team building VideoDB.
featured
latest
Operations
Stop paying the TCP/TLS tax: connection pooling for Python services
Replacing bare requests calls with a shared session is not glamorous. It is table stakes for production services that call the same hosts repeatedly.
Frontend Performance
One CORS header that removed hundreds of milliseconds from repeated API calls
A practical note on preflight caching, custom auth headers, and why OPTIONS requests quietly dominate app latency on slower connections.
Operations
A Postgres backup is not real until you restore it
A short operational checklist for dump validation, restore drills, row checks, and the habits that make backups useful under pressure.
Build Systems
How we made Python ML containers smaller and faster to deploy
Slim bases, same-layer cleanup, CPU-only PyTorch, pinned wheels, and the unglamorous work of keeping deploys fast.
dispatch
Newsletter / Video agents · model watch · VideoDB updates
The VideoDB Dispatch #001
A curated note on agentic video, generated visual media, and what we shipped at VideoDB.
research
Papers by us
Research at the edge of video, vision-language models, and agents.
Published work and preprints from the VideoDB team. More papers will be added here as they are shared.
arXiv preprint arXiv:2604.11177 / 2026/4/13
Do Thought Streams Matter? Evaluating Reasoning in Gemini Vision-Language Models for Video Scene Understanding
Benchmarks how internal reasoning traces affect video scene understanding in Gemini models, including where quality gains plateau and how reasoning budgets influence hallucination.
Read on arXivarXiv preprint arXiv:2502.06445 / 2025/2/10
Benchmarking Vision-Language Models on Optical Character Recognition in Dynamic Video Environments
Introduces an open-source benchmark for evaluating VLMs on OCR tasks in dynamic video environments across 1,477 manually annotated frames.
Read on arXivplaygrounds
VideoDB Playgrounds
Small projects and experiments from the team.
Utilities, creative experiments, research notebooks, and independent projects built by people on the VideoDB team.
By Ashu
Chitrakāvya
A research archive that rebuilds Sanskrit picture-poems as computational artefacts, connecting śloka, geometry, knight's tours, palindromes, and early algorithmic thinking.
Visit project →
By Ashu
Emergence
A blog on multi-agent systems, collective intelligence, alignment, swarm behavior, and computational models of consciousness.
Visit project →
By Ashish
Cool QR Codes Studio
A no-login QR code studio for turning links or text into customizable, export-ready QR codes with scanability checks.
Visit project →
By Ashish
makememes.site
A code-first video meme generator built around reusable templates, a meme bank, custom text overlays, and VideoDB-powered source videos.
Visit project →