Skip to content
View divyanshAgarwal123's full-sized avatar

Highlights

  • Pro

Block or report divyanshAgarwal123

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
divyanshAgarwal123/README.md
Divyansh Agarwal — AI systems, security, and creative engineering. Animated signal diagram.

Engineer by training. Builder by habit.

I study Computer Science (AI & ML) at Symbiosis Institute of Technology, Pune and build at the intersection of agent systems, security, developer tools, and interactive experiences.

Explore the work · How I build · Connect


The short version

I like projects where a useful idea meets a hard boundary: a model needs evidence, an agent needs permission, or a deployment pipeline needs a reason for its recommendation. My work spans local security experiments, governed agent actions, CI tooling, and visual web experiences.

curiosity  →  prototype  →  test the failure mode  →  make the evidence readable  →  iterate
  • Now: Campus Lead, OpenAI Student Collective at SIT Pune; building peer workshops and hands-on AI sessions.
  • Previously: Software Development Engineer Intern at SMCS; freelance web developer for FIVE R.
  • Based in: Pune, India · B.Tech in Computer Science (AI & ML), 2024–2028.

Selected work

Animated project map: MUTE observes isolation boundaries, Railgate constrains agent checkout, and EcoCI improves CI workflows.

Python · security testing · evidence design

Models shared mutable resources between two runs, then tests predicted transfer paths with isolated writer and reader processes. The repository includes negative controls, 32-trial archived evidence, integrity checks, structured JSON, and a standalone visual report. It is a local synthetic prototype; its observations do not claim a live-service vulnerability.

TypeScript · MCP · SQLite · Razorpay test mode

Puts a deterministic policy layer between an AI agent and checkout. Atomic spend reservations, single-use human approval, a server-owned catalog, and an audit ledger keep purchases bounded and explainable. The repo includes automated tests and a test-mode provider flow.

Python · GitLab/GitHub workflows · developer tooling

Analyzes pipeline configuration and run signals across security, flaky tests, delivery metrics, runner cost, and estimated carbon impact. It can produce fix recommendations and optimized workflow candidates through a CLI, automation, and editor surfaces.

More of the build log
  • FIVE R: Designed and built a cinematic product website for a sustainable-materials brand using Next.js, TypeScript, React Three Fiber, Three.js, GSAP, and Blender.
  • SMCS operations platform: Expanded a bill-of-materials tracker into a four-division internal tool with React, Node.js/Express, SQLite, authentication, and spreadsheet import/export.
  • ClaimTrail: A local review workbench that places an agent's completion claims beside recorded actions and artifact checks. Its demo data is synthetic.

How I build

Layer Tools I use What I care about
Languages Python, TypeScript, JavaScript, C, C++ Clear interfaces and reproducible runs
Product React, Next.js, Node.js/Express, Three.js, GSAP Interfaces that make complex systems legible
AI & data scikit-learn, TensorFlow Lite, OpenCV, YOLO, MCP Useful models with explicit limits
Systems SQLite, MongoDB, MySQL, GitLab CI/CD, GitHub Actions Audit trails, tests, and controlled automation
Physical & visual Arduino, ROS, Blender, Unity AR Prototypes you can see and interact with

Outside the code editor

At SIT Pune, I help run the OpenAI Student Collective as a campus lead, organizing hands-on learning around AI tools and projects. Before university, I led an Atal Tinkering Lab and mentored more than 100 students through electronics and robotics prototypes. I also received the INSPIRE-MANAK Award for a towing alert prototype and placed third at 2 Fast 2 Hack with a team emergency-assistance app.

Connect

If a project here sparked a conversation, reach me on LinkedIn or email. Browse the repositories for code and project-specific setup notes.

Build it. Test the boundary. Show the evidence.

Pinned Loading

  1. claimtrail claimtrail Public

    Local evidence workbench for task-completion claims in collaborating agent groups; synthetic demo and bounded AI-assisted research overview.

    Python

  2. ecoci ecoci Public

    CI/CD analysis and optimization across security, reliability, cost, and estimated carbon impact.

    Python

  3. LED-MASK- LED-MASK- Public

    Arduino-based LED mask with animated eyes and smile

    C++

  4. mute mute Public

    Pairwise containment tests for shared-state leakage between AI-agent runs

    Python

  5. railgate railgate Public

    Governed agentic checkout with atomic spend caps, human approval, MCP tools, and an audit ledger.

    TypeScript

  6. signalpost-company-research-agent signalpost-company-research-agent Public

    Evidence-first Norwegian company research agent with exact-entity checks, sourced claims, and refreshable profiles

    Python