This renderer optimization is deployed to production. I replaced serial JavaScript frame encoding with browser-native recording, moved composition into a prerender cache, and removed repeated asset work. In the matched staging benchmark, render time fell from 65 to 28 seconds, with smaller output and bounded per-job memory. Quality checks and independent rollback switches remain in place.
I’m architecting a QA process that follows a Bitbucket pull request from its versioned Confluence PRD through test design, execution, human UAT and production regression. Six specialist agents handle repeatable work; people approve tests, releases and production responses. The proposal defines the gates, evidence and failure paths for a one-team pilot.
I architected the PostgreSQL model and built the DRF API layer for live classes, test series, video lectures, and student analytics. I standardized EC2 deployment and led the engineering support practices needed to maintain the platform.
Siddharth Juyal built this PIM ingestion rebuild at Tyroo; I was a collaborator. The two-pass Polars pipeline processes multi-million-product catalogs within 2 GB of RAM, with field-level parity checks. This study explains his rebuild and the related team work.
Follow my production-deployed Renderer work from a slow frame-encoding loop to native recording and correctness checks. The 65-to-28-second result is from matched staging runs.
Follow the systemOwned study 01 / Performance engineering
Constraint / 01
A six-second video took 65 seconds to render.
The old page serialized, decoded and software-encoded every frame before assembling an all-keyframe WebM. Repeated asset preparation and a 112 MB per-job memory leak added separate latency and capacity costs.
180 serial frames for six seconds at 30 fps
All-keyframe output around 25 MB
+112 MB retained memory per job
01SVG composition
02WebP + WebMWriter
03Base64 transfer
Conceptual sequence · based on the project narrative
My work connects architecture, implementation, and production support. I care about how a system behaves after it ships.
Vaibhav LehraDelhi NCR, India
I turn complex backend problems into systems people can rely on.
With 6+ years of experience across AdTech and EdTech, my work spans renderer performance, QA systems architecture, and learning-platform backends.
At Tyroo, I implemented the video renderer optimizations featured here and am architecting a QA automation proposal with human approval gates. Earlier, at Myclassroom, I built the PostgreSQL model and DRF APIs, standardized deployment, and led engineering support.
I also mentor a 10-engineer team, review agent-assisted development, and work with product, QA, and ad-ops to turn technical decisions into working systems.
Where it started
B.Tech in Electrical and Electronics Engineering
Guru Tegh Bahadur Institute of Technology, GGSIPU · 2018
Work on distributed services across Dynamic Product Ads and product catalogs: renderer optimization, schema and API design, Docker / Kubernetes deployment, CI, monitoring, and production support. Architect a human-gated AI QA process, mentor a 10-engineer team, and lead reviewed LLM-assisted development workflows. Shipped a React / TypeScript integration end to end.
Architected the PostgreSQL learning-platform model and built DRF APIs for React web and mobile clients. Standardized EC2 deployment with NGINX and Gunicorn; built and led engineering support, mentored engineers, and established production triage and escalation.
Linked tools lead to a study of their use. Additional experience includes React, Redux, TypeScript, OAuth 2.0, JWT, Elasticsearch, and AI-assisted development. Terraform: working knowledge. GCP: transferable cloud experience.
04 / Start a conversation
Let’s build what comes next.
A renderer that needs to move faster. A backend ready for its next stage. A QA process that needs a clear path from requirements to release.
If you’re building systems like these, or looking for a senior Python engineer, I’d like to hear from you.
