Hi, I'm Atri. I build products
and the infra under them.

Took the AI features from "cool demo" to something the team actually relies on.

• Built the prototype for an AI Interview Assistant — LLM resume parsing, contextual interviewing, evaluation — and pulled the internal rollout in by two months.
• Designed the PromptBuilder pipeline (backend metadata + a learner's full-stack project data + LLMs) and cut a 6-week timeline down to 2.
• Sat through 10+ learner interviews, found the real UX friction in activity design, and the redesign moved completion up 24%.
• Owned session-level NPS for mentor-led cohorts and kept it at 8.7 on average.
• Built a root-cause audit tracker across 1,500+ automated support tickets so the team argued about data instead of vibes.

Promoted twice. Mostly curriculum, autograders, and the deploy pipeline.

• Conceptualized and launched India's first Advanced SDET program — 13 modules, 92% completion, 80+ NPS.
• Dug through behavioral data (SQL + Python) for 20+ users, found people misreading optional content, and redesigned the UI to fix it.
• Put Kafka + Docker hands-on labs into the curriculum so the cloud stuff stopped being theoretical.
• Wrote Python + Docker autograders for 30+ projects — grading got 30% faster and feedback became real-time.
• Scripted away ~80% of the manual Discourse content ops across 1,500+ articles.
• Set up Jenkins-based deploys for backend services on Cloud Run + Docker.

An AI JEE-prep app, built from nothing with two hands and a cloud bill.

• Bootstrapped the whole thing on Supabase, GCP and AWS — frontend to infra.
• Wrote and shipped a modular AI Service (FastAPI + PostgreSQL on AWS EB) that orchestrated between the LLMs and the platform.
• Used low-temperature, multi-model validation heuristics to keep hallucinations down across 3 evaluation pipelines.
• Wired CI/CD with Amplify + Cloud Run and cut deploy time 80% with Docker layer caching.
• Built the Next.js landing page; bounce rate dropped 23%.