AI collaboration profile

Software builder using agents to plan, ship, and verify.

A compact public view of sustained AI-assisted engineering work: planning, shipping, verifying, debugging, documenting, and publishing without exposing project identities.

Current focus AI-native product engineering and agent workflows
Core strengths Architecture, implementation, debugging, verification
Location India / IST
Links Add website, GitHub, LinkedIn, or contact

Mandar Badve

AI-assisted software builder Product engineering Agentic workflows Systems, apps, and publishing

Software builder using AI agents as a serious engineering interface: plan the work, execute in tight loops, verify outputs, and turn useful systems into shareable artifacts.

Primary mode Builder, reviewer, debugger, publisher
Work style Spec-driven, test-aware, privacy-conscious
Public summary 490 Codex threads across 56 active days

Activity Snapshot

490Codex threads indexed
56active days observed
38distinct workspaces
46k+tool events in rollout logs

What Shows Up Most

Architecture
43%
Engineering
18%
Publishing
12%
Verification
8%
Debugging
7%
AI systems
7%
Infrastructure
6%
Operating pattern

The strongest signal is structured execution: plans, phases, specs, implementation passes, verification loops, and review.

Typical work
  • Full-stack and app implementation
  • Debugging and repair of failing flows
  • Test, CI, and browser verification
  • Agent, AI, data, and deployment workflows
  • Documentation, publishing, and public artifacts

Operating System

The profile is less about one project and more about a repeatable way of working with AI agents: clarify the shape of the work, execute in small passes, and verify before calling anything done.

01
Spec the work

Turn rough ideas into scoped plans, explicit constraints, and concrete acceptance checks before implementation starts.

02
Build in loops

Use agents for implementation, review, debugging, documentation, and test feedback instead of one-shot generation.

03
Verify the result

Prefer runnable checks, browser inspection, CI-style validation, and direct artifact review over confidence alone.

AI Workbench

Metadata-derived tool signals from Codex interactions, keyword matches, and safe tool-call names across planning, coding, testing, data, deployment, and publishing.

Codex TypeScript GitHub React Expo Playwright Supabase / Postgres Vercel Cloudflare here.now

Agent Stack

Metadata-derived plugin and skill signals are tracked separately from languages and broad tools, then grouped by what they help accomplish.

Mobile and native
Expo

Mobile and React Native surfaces appear as the strongest plugin-layer signal in the current scan.

Workflow layer
Superpowers

Planning, test-driven development, systematic debugging, review, and verification-before-completion patterns.

Data and backend
Supabase

Postgres, migrations, schema work, authentication surfaces, and database-backed application workflows.

Browser QA
Browser / Playwright

Browser checks, screenshots, interaction testing, and e2e verification show up as repeated workflow signals.

Repository and CI
GitHub

Repository review, CI language, and GitHub-oriented workflows are present in the safe metadata corpus.

Cloud and deploy
Vercel / Cloudflare

Workers, Wrangler, deployment, hosting, routing, and web platform operations around public artifacts.

Conservative keyword scan

Match known plugin and skill names in thread titles, first user messages, prompt history, and safe rollout text.

Structured event extraction

Parse tool namespaces, MCP calls, and skill references when the session records expose them without copying raw private content.

Directional, not exhaustive

Counts should represent observed stack signals, not a complete transcript or proof of every plugin loaded in a session.

Languages

Metadata-derived language signals from prompts, thread metadata, and safe tool names. Languages only appear here when the scan finds enough signal to publish them.

Primary
TypeScript / JavaScript

Frontend, full-stack app work, automation scripts, browser verification, and UI publishing workflows.

Used
Python

Data processing, local analysis scripts, workers, automation, and support tooling around AI-assisted workflows.

Current Direction

Directional themes ranked from recent and repeated metadata signals, without naming private products, customers, repositories, or internal workflows.

Exploring
Agentic development systems

Workflows where agents help plan, implement, inspect, and maintain software across multiple sessions.

Sharing
Useful public artifacts

Profiles, pages, prototypes, docs, and demos that can be safely published without leaking private project context.

Building
AI-native product interfaces

Product experiences that use AI as part of the core workflow, not a decorative layer added after the fact.

Evidence Trail

The public numbers are aggregate signals from local agent metadata. They show operating rhythm and tooling depth while keeping private project context out of the page.

Observed window
01
Jan 21, 2026 First observed Codex activity in the local metadata set.
02
Jun 1, 2026 Latest observed Codex activity included in this profile snapshot.
03
56 active days Unique UTC calendar days with observed AI-assisted work.
Source ledger
490
490 state threadsThread rows from the local Codex state database.
492
492 rollout filesSession JSONL files used for broad event counting.
464
464 session-index recordsRows from the local Codex session index.
82
82 prompt-history recordsPrompt-history rows counted for aggregate activity only.
331
331 ChatGPT blobs counted onlyLocal binary conversation blobs were counted, not decoded or analyzed.
Public safety boundary
  • No raw prompts
  • No repository paths
  • No client or company names
  • No private file contents

Connect

Add exact public links when ready. Until then, this section keeps the page ready for publishing without inventing contact details.

WebsiteAdd URL
GitHubAdd handle
LinkedInAdd profile
EmailAdd public email

AI collaboration profile generated from Codex metadata: 490 state threads, 492 rollout files, 464 session-index records, and 82 prompt-history records, with 331 ChatGPT conversation blobs counted as part of the broader activity snapshot.

Built with the open-source Codex Profile Skill.