India | Building agents that actually do the thing | Siemens by day, native + agent tools by night
Engineer building local-first AI tools, agent runtimes, native macOS apps, and developer infrastructure.
I care about agents that are debuggable, evaluable, and useful outside demos.
- Agent runtimes with orchestration, evaluation, debugging, and evolution loops
- Local-first dev tools for PDFs, Jira-style tracking, markdown, dashboards, and code comprehension
- Native macOS apps that stay fast, private, and battery-friendly
- VS Code extensions that turn the editor into a thinking surface
- Personal AI infrastructure around Kayo, my vault-backed assistant for memory, scheduling, web fetches, finance, and Telegram workflows
| Project | What it does | Stack / Focus |
|---|---|---|
| agent-ville | Evolves populations of prompt-tool-strategy genomes through scoring, mutation, and selection | Python, JS, agent evaluation |
| ForrestRun | Runtime for designing, debugging, and teaching agents | Agent runtime |
| atlas | Native macOS PDF reader with AI-generated knowledge maps and source-linked concept graphs | Swift, macOS, AI reading |
| gitstatus | Lightweight menu bar Git repo monitor with near-zero idle overhead | AppKit, macOS |
| flowDE | Graph-based Python code comprehension inside VS Code | TypeScript, static analysis |
| kayo | Personal AI assistant wired into memory, scheduling, finance, web fetches, and Telegram | AI assistant infrastructure |
-
🧬 agent-ville
Agents that evolve through Darwinian selection. Spawns populations of prompt-tool-strategy genomes, scores them, mutates the survivors, and keeps the best performers.
Focus: agent evaluation, evolutionary loops, pure stdlib + vanilla JS. -
🌲 ForrestRun / agentic-runtime
Runtime for designing, debugging, and teaching agents. Built around making agent behavior observable instead of mysterious.
Focus: agent runtime, debugging, orchestration. -
🦞 closed_claw
Framework for automated agent creation, deployment, orchestration, and storage. A growing village of persistent agents.
Focus: agent creation, deployment, orchestration. -
🌐 mycompagent
Agent that drives your browser through the Playwright CLI.
Focus: browser automation, agent control, Playwright. -
🏗️ AI-System-Engineer
Agent that approaches problems like a systems engineer.
Focus: systems thinking, architecture, AI-assisted engineering. -
🤖 kayo
My personal AI assistant. Vault-backed knowledge ingestion, finance dashboard, scheduler, web fetcher, Telegram bridge, and a unified Copilot/Claude Code identity.
Focus: personal AI infrastructure, memory, automation.
-
🗺️ atlas
Native macOS PDF reader with a live, AI-generated knowledge map. Creates force-directed concept graphs linked back to source passages, with cross-document correlation built in.
Focus: Swift, macOS, PDF intelligence, knowledge graphs. -
✍️ markit
Native macOS app for annotating Markdown. Highlight, comment, and export without modifying the source files.
Focus: Swift, native Markdown workflows, no Electron. -
🌳 gitstatus
Tiny menu bar app that scans~/Documentsfor Git repositories and shows branch + working-tree state.
AppKit, 0% idle CPU, single-digit MB RSS.
Website: gitstatus.tech
Focus: AppKit, Git tooling, lightweight native utilities.
-
🕸️ flowDE
Graph-based code comprehension for Python inside VS Code. Static analysis, data-flow tracking, and interactive flowchart exploration.
Focus: VS Code, Python analysis, code graphs. -
💬 mdcomments
Inline, thread-based comments on rendered Markdown inside VS Code.
Focus: Markdown workflows, VS Code extension development. -
📋 jira
Local, file-backed Jira-style project tracker. No cloud, no login, no SaaS bill.
Focus: local-first project management. -
🧪 sample_backends
Runnable sample servers for local development, integration testing, and demos. Includes FastAPI/Flask examples for stores, payments, file upload, search, messaging, and OTP auth.
Focus: backend demos, API testing, local integration.
- 🎓 system-design-tutor
Claude Code skill for running an end-to-end system design course. Anchored to DDIA and the System Design Primer, with real-world incidents, runnable exercises, and mock interviews.
Focus: system design, Claude Code skills, structured learning.
- 📱 scroller
Agent that drives your phone for you. Android first.
Focus: mobile automation, Android, agent control.
-
📄 idolovepdf
Local PDF toolkit with CLI and browser UI. Merge, split, rotate, compress, watermark, number, reorder, and convert PDFs without uploads or cloud processing.
Focus: local-first PDF tooling. -
📊 GlanceOS
Tile-based dashboard for Raspberry Pi, Linux, and macOS. Glassmorphic draggable widgets, real-time WebSocket data, and lofi ASCII vibes.
Focus: dashboards, Raspberry Pi, real-time widgets. -
💚 wa-puppet
WhatsApp Web automation in Python + Playwright with a JSON rule engine.
Focus: WhatsApp automation, Playwright, rules engine. -
🗃️ wa-data
Pipe WhatsApp group messages into SQLite using a tiny Node bot and FastAPI backend.
Focus: message ingestion, SQLite, FastAPI.
Earlier Work
-
✏️ agentic_article_writer
Multi-step article-writing agent that adapts to different article styles. -
🩺 doctor_reporting
Agent flow for auto-generating consultation reports from patient discussions. -
💧 AquaTwin
AI-driven control layer for a water-management digital twin. The pattern generalizes to other digital-twin systems. -
🧮 neural_network_numpy
Neural network built only with NumPy. No TensorFlow, no PyTorch. -
🎸 math_guitar_VLM
Vision + image-processing-based pedagogy for math and guitar. Generalizes to learning workflows beyond those two domains. -
📝 notes_me
Android second-brain app for personal notes and memory.
I like tools that are:
- Local-first: useful without a cloud dependency
- Inspectable: easy to debug, trace, and understand
- Fast enough to disappear: no heavy UI tax, no idle battery drain
- Built around real workflows: not demos, not toy wrappers
- Composable: small pieces that can be wired into larger systems
Open to collaborating on agent runtimes, local-first AI tools, native apps, developer infrastructure, and serious product ideas.