Skip to content
AP
AP

Generated from the same data as this site — it cannot drift.

Plain text (ATS)PDF

Anshul Patil

Backend & Systems Developer | GSoC @ BRL-CAD

Mumbai, Maharashtra · anshulpatil1022@gmail.com · GitHub · LinkedIn

Summary

I build distributed systems and AI infrastructure: open-source geometry processing, document intelligence pipelines, and production-grade developer tools. Currently pursuing B.Tech at IIIT Surat and contributing to BRL-CAD via Google Summer of Code.

Education

B.Tech, Indian Institute of Information Technology, Surat

2024 – 2028

Experience

Selected Contributor, Manifold Subproject — Google Summer of Code, BRL-CAD

Present
  • ▸Selected to improve BRL-CAD's Manifold C++ geometry processing library through CI reliability and benchmarking infrastructure.
  • ▸Designed cross-platform determinism checks using fixed mesh cases, canonical artifacts, SHA256 comparison, and clear mismatch reporting.
  • ▸Planned Linux Clang ASan+UBSan testing and PR/weekly benchmark workflows with base-vs-head comparison, JSON history, and dashboard trend visualization.

Open Source Contributor, AI Systems — Extralit Open Source

Dec 2025 – Present
  • ▸Contributed to an AI document intelligence platform supporting OCR parsing, dataset workflows, structured extraction, and annotation pipelines.
  • ▸Merged 10+ pull requests across ingestion pipelines, dataset configuration, frontend components, validation logic, and test coverage.
  • ▸Fixed multi-step workflow persistence issues and added structured table-based dataset inputs with validation and UI integration.

Software Engineer Intern — Bluestock™

Dec 2025 – Feb 2026
  • ▸Improved system reliability and performance across user-facing financial features.
  • ▸Wrote clean, well-tested code and performed debugging across the codebase.
  • ▸Collaborated remotely with the team over a 3-month internship engagement.

Software Developer Intern — Techvisio Design

May 2025 – Aug 2025
  • ▸Built a full-stack analytics dashboard using React, Django REST Framework, and SQL APIs to process 10K+ daily user events.
  • ▸Integrated AWS S3 pipelines for 300K+ monthly records and reduced backend API latency by 40% through query optimization and caching.

Products

Stratum — AI code intelligence, deployment risk, incident correlation

  • ▸Reviews GitHub pull requests with typed findings, risk scoring, per-file summaries, review modes, async jobs, reruns, comparisons, and exports.
  • ▸Analyzes deployment batches to detect semantic conflicts between PRs before they are shipped together.
  • ▸Builds a live architecture-drift view from accumulated review data, highlighting risky modules, boundary erosion, and refactor pressure.

docRAG v3 — PDF RAG, OCR, semantic search, research GraphRAG

  • ▸Processes PDF uploads asynchronously with FastAPI, Celery, Redis, OCR, sentence-transformer embeddings, and Qdrant vector search.
  • ▸Supports natural-language chat over documents with citations that point back to source document chunks and pages.
  • ▸Includes an Angular frontend for upload, task status tracking, health checks, recent tasks, and document query workflows.

Selected projects

AI Pull Request Reviewer — FastAPI, React, TypeScript, GitHub API, OpenRouter

Full-stack AI pull request reviewer that fetches GitHub PR diffs, runs LLM-assisted risk analysis, persists review history, and returns structured findings, suggestions, per-file summaries, and exportable review reports. Open-source and self-hostable; its architecture became the foundation Stratum was built on.

sect_scrape — Python, Playwright, BeautifulSoup, SQLite, OCR

Research-grade scraper for Gujarat eCourts disposed CRMA/JMFC cases, focused on CrPC sections 436, 437, 438, and 439. It automates the eCourts portal with Playwright, handles CAPTCHA/OCR support, stores structured metadata in JSON/JSONL/SQLite, and preserves source HTML/PDF orders for verification.

Faulty Node Detection — NS-3, Python, anomaly detection, HTML dashboard

Four-stage faulty-node detection pipeline for simulated networks. It builds a 12-node NS-3 topology, injects packet-loss/high-delay/low-bandwidth faults, extracts FlowMonitor features, detects anomalies with z-score/IQR/composite scoring, and explains results through an interactive HTML dashboard with an LLM diagnostic agent.

Codeforces — FastAPI, React, TypeScript, Codeforces API

Modern full-stack Codeforces dashboard for contest profiles, rating visualizations, performance insights, and problem browsing. The backend uses FastAPI with rate limiting and caching around real Codeforces data; the frontend uses React, TypeScript, Tailwind, React Query, and Recharts.