I'm a software engineer who builds and evaluates AI and machine-learning systems — and cares about making them reliable, not just demos. By day I'm on Oracle's MySQL & HeatWave team, working on CI/CD automation and the rigor of evaluating things against clear standards. Outside that, I build applied-AI and full-stack projects: retrieval/RAG systems, LLM evaluation tooling, and the engineering that gets them shipped.
- 🔭 Focused on applied AI — RAG, LLM evaluation, and ML systems, backed by solid Python/TypeScript engineering
- 🌱 Always learning — lately around LLM evaluation, retrieval, and agentic workflows
- 💜 Committed to growing the representation of women in STEM
- 📫 Reach me at vermaapoorva0510@gmail.com
All highly competitive, national-level selections.
- D. E. Shaw — DESIS Ascend Educare Fellow — 1 of 35 selected from 6,200+ applicants across India
- JPMorgan Chase — Quantitative Research Mentee — 1 of 35 students across India
McKinsey & Company — NextGen Women Leaders — top 50 women in India
Microsoft — Engage Mentee (2022)
Oracle — MySQL & HeatWave Release Engineer — cut integration issues ~30% across 100+ automated test suites; integrated OpenAI Codex into the release workflow; CI/CD across 5 time zones
Handshake AI — Project Dynamo (contract) — authored frontier-agent evaluation tasks (Terminal-Bench 2 / Harbor)Oracle — Software Engineer Intern — real-time multimedia + video-annotation tools for Oracle Content Management
- 33 merged PRs across 10 external projects — Supabase (×3), sqlfluff (×2) ⭐, PostHog, Falcon, pyparsing, pygments, mne-python, PapaParse, jsonpickle, twig.js
- 18 accepted answers in GitHub Discussions ⭐ across 15 repos — FastAPI/SQLModel, DuckDB, ripgrep, Rich/Textual, Starlette, spaCy, Apache Airflow, dbt-core, Haystack, turso ⭐, and more
AI reliability & evaluation — my core focus: making LLM systems dependable, not just demos.
- langchain-shannonbase —
Python · LangChain · MySQLA LangChainVectorStorefor MySQL 9's nativeVECTORtype — RAG on ShannonBase, self-hosted MySQL, or HeatWave, no separate vector DB. Fills a real ecosystem gap. - atlasai-mcp-bridge —
TypeScript · Atlassian ForgeAn MCP bridge that brings Jira, Confluence & JSM into Claude, Cursor, and any MCP client — OAuth 2.1, Forge SQL, audit logging, React admin UI. - rag-eval-harness —
PythonRAG over research papers with a reproducible retrieval-evaluation harness (DeepEval + a custom citation-faithfulness metric). - claude-code-toolkit —
ShellSkills, subagents, and hooks that make Claude Code follow good engineering practices by default. - dra-industry-4.0-assessment-platform —
Next.js · PostgreSQLA digital-readiness platform that scores an organization's Industry 4.0 maturity and surfaces actionable insights.
- Published ML research — "Crime Detection System using Face Recognition", Indian Journal of Applied Research (2022) — 88%+ accuracy, sub-4s inference
- Oracle Cloud Infrastructure 2024 Certified Foundations Associate
Building and evaluating AI systems that actually ship — and growing women's representation in STEM along the way.




