class PaarthGala:
role = "Founding AI Engineer @ Digital Crew"
focus = "AI agent infrastructure"
building = [
"MCP servers with repo memory",
"Max Agent + Max MCP",
"Claire Agent + Claire MCP",
"VPS/Docker agent workflows",
]
recent_work = {
"Digital Crew": "Founding AI Engineer · May 2026 - Present",
"Shipyard": "AI Infrastructure Intern · Mar 2026 - May 2026",
"Open Source": "MCP + developer tools contributor",
}
# collab portal: paarthgala1@gmail.comI build AI agent infrastructure - MCP servers, repo-memory systems, VPS/Docker deployments, backend automation, model-routing systems, integrations, workflows, and developer tools that make agents useful in real products.
current_focus/
├── MCP servers and agent tools
├── repo memory for AI coding agents
├── VPS and Docker agent deployments
├── AI agent integrations and workflows
├── model routing and telemetry
└── backend automation for real users🚀 Open current build mode
mode: shipping useful AI agents
status: building in public
runtime: caffeine + curiosity + real user workflows
priority: make agents remember context, use tools, and survive production🧠 What I can nerd out about
MCP servers → tools, memory, permissions
AI agents → Max, Claire, Orchestrator, repo-aware coding agents
backend systems → auth, queues, Docker, VPS, diagnostics
product infra → model routing, telemetry, latency, cost controls
devtools → CLIs, JSON output, workflows for humans + agents🤝 Good collab quests
bring me:
├── an agent idea that needs real tools
├── an MCP server that should exist
├── a workflow that is too manual
└── an AI product that needs to become production-gradePAARTH_GALA/
├── main_quest/
│ ├── Build reliable AI agents
│ ├── Create MCP servers for real workflows
│ ├── Make agents remember context
│ └── Turn AI demos into production systems
│
├── production_quests/
│ ├── Digital Crew Agent Systems
│ │ ├── Max Agent
│ │ ├── Max MCP
│ │ ├── Claire Agent
│ │ ├── Claire MCP
│ │ └── Orchestrator workflows
│ │
│ ├── Shipyard Model Router
│ │ ├── agent telemetry
│ │ ├── cost-aware model routing
│ │ ├── latency analysis
│ │ └── self-improving AI infrastructure
│ │
│ └── NSE News-to-Telegram Automation
│ ├── financial news ingestion
│ ├── signal filtering
│ └── Telegram delivery pipeline
│
├── open_source_quests/
│ ├── OpenSRE → AI SRE + delivery hardening
│ ├── aden-hive → developer onboarding docs
│ ├── Lamatic → AgentKit automation config
│ ├── hstack → Hermes skill catalog
│ ├── PR Context MCP → repo memory for agents
│ └── Quant Brain MCP → finance analysis over MCP
│
└── side_quests/
├── writing in public
├── AI infra research
├── startup experiments
└── learning backend systems deeply🧠 Boss 01: Stateless AI Coding Agents
Most AI coding agents forget repo history, review preferences, architectural decisions, and team patterns.
My attack: I built GitHub PR Context MCP to give agents access to historical PRs, review threads, comments, and repo-specific context.
Why it matters: AI agents become more useful when they understand how a team actually ships code.
💸 Boss 02: Expensive AI Workflows
Not every task needs the most expensive model. Some tasks need speed. Some need reasoning. Some need more context.
My attack: At Shipyard, I worked on model routing and telemetry architecture for a self-improving AI cost-optimisation system.
Why it matters: The future of AI products is not just better prompts - it is routing, telemetry, feedback loops, and cost-aware execution.
⏳ Boss 03: Long-Running Agent Jobs
AI workflows often run longer than serverless limits. If they block the request, the user sees a timeout.
My attack: At Digital Crew, I worked across Max Agent, Max MCP, Claire Agent, Claire MCP, and Orchestrator workflows to connect AI agents with real product integrations.
Why it matters: Production AI needs orchestration, not just model calls.
⚡ Boss 04: AI Product Latency
Agents become painful when they rebuild huge context every turn or block streaming responses.
My attack: I analysed Paula AI latency bottlenecks and proposed prompt caching, streaming response optimisation, and better static/dynamic context separation.
Why it matters: Good AI products must feel fast, not just smart.
🛡️ Boss 05: Fragile Production Integrations
APIs fail, serverless environments miss system libraries, secrets leak into logs, and error bodies are not always JSON.
My attack: I contributed reliability hardening across Digital Crew and OpenSRE: agent diagnostics, auth paths, token redaction, non-JSON error handling, and safer delivery paths.
Why it matters: Production systems fail in boring ways. Good engineers handle the boring failures before users see them.
🤖 Digital Crew Agent Systems - Max, Claire, MCP, and Orchestrator workflows
Founding AI Engineer work across Digital Crew, focused on production AI agent infrastructure for Max, Claire, MCP servers, Orchestrator workflows, integrations, and deployed backend automation.
product workflow
→ Max Agent
→ Max MCP
→ Claire Agent
→ Claire MCP
→ Orchestrator workflow
→ integrations
→ deployed automationWhat I worked on:
- Max Agent workflows for product-facing AI automation
- Max MCP server work for structured agent tools and integrations
- Claire Agent workflows for AI-assisted business operations
- Claire MCP integrations for agent-accessible product capabilities
- Orchestrator workflow wiring across backend services, agent tools, and external integrations
- VPS and Docker-based deployment work for reliable long-running agent services
- Backend diagnostics, auth, and failure handling for production AI workflows
Stack: Next.js · TypeScript · Supabase · VPS · Docker · AI agents · Integrations · Workflows · MCP
🚢 Shipyard AI Model Router - model routing + telemetry architecture
Worked on AI infrastructure for Shipyard, an AI-native agile project management platform, from March 2026 to May 2026.
What I worked on:
- Cost-aware model routing framework
- Structured telemetry schema
- Agent execution event loop
- Task/domain classification
- Outcome and latency tracking
- Paula AI latency analysis
- Prompt caching and streaming optimisation design
Core idea: Route each AI task to the cheapest model that can still complete it reliably.
🧩 GitHub PR Context MCP - repo memory for AI coding agents
AI coding agents need institutional memory.
This MCP server fetches repository PR history, comments, review threads, and team patterns, then turns them into searchable context materials for coding agents.
Use cases:
- Historical code review context
- Team-specific coding rules
- Grounded code generation
- Repo-aware AI assistant workflows
📦 hstack - portable skill catalog for AI coding agents
A skill catalog that helps AI coding agents deploy self-hosted Hermes agents and wire them into external services.
What it supports:
- Hermes deployment
- 70+ integrations
- SSH-first workflows
- Dry-run preview
- Rollback-safe execution
- Skills for AI coding agents like Codex, Claude Code, Cursor, Gemini CLI, and Hermes
📈 Quant Brain MCP - finance intelligence over MCP
MCP server for stock analysis, backtesting, portfolio optimisation, sector intelligence, and chart generation.
Capabilities:
- US + India ticker support
- Strategy backtesting
- Portfolio optimisation
- Sector rotation analysis
- 150+ indicators
- Chart pack generation
🧯 OpenSRE Contributions - AI SRE reliability hardening
Contributed to AI SRE tooling by hardening Slack and Discord delivery paths.
Work included:
- Token redaction
- Safer handling of non-JSON API failures
- Defensive exception handling
- Unit test coverage for delivery edge cases
🔌 killport - Docker-native port killer for humans and agents
A fast port-killing developer tool with machine-readable JSON mode for AI agents and CI workflows.
Why it matters: Agents waste tokens and make mistakes when CLI output is messy. Tools should return structured output.
⚙️ Digital Crew Technologies - Founding AI Engineer
Founding AI Engineer building production AI agent infrastructure across Digital Crew, Max, Claire, MCP systems, Orchestrator workflows, backend integrations, and deployed automation.
Period: May 2026 - Present
Highlights:
- Worked on Max Agent, Max MCP, Claire Agent, and Claire MCP systems for production AI workflows.
- Built and connected MCP-style tools, backend APIs, and integrations so agents could perform useful product actions instead of staying as chat-only flows.
- Worked on Orchestrator workflow wiring across agents, services, integrations, persistence, and diagnostics.
- Contributed VPS and Docker deployment work for AI agent services that need reliable runtime behavior beyond simple request/response execution.
- Built production diagnostics, auth paths, fallback handling, and backend automation for agent workflows across Max and Claire.
🚢 Rowan Cognitive / Shipyard - AI Infrastructure Intern
Worked on AI infrastructure for Shipyard, focusing on model routing, telemetry, cost optimisation, and agent performance.
Period: Mar 2026 - May 2026
Highlights:
- Designed structured telemetry for the full AI agent execution loop, including user intent, prompt construction, model execution, token usage, latency, call type, and final outcome.
- Helped design a cost-aware model routing framework to choose model tiers based on domain, effort, context needs, latency sensitivity, and historical outcome signals.
- Analysed agent behaviour, production run data, and workflow patterns to identify cost-impact areas, routing opportunities, and quality-risk tradeoffs.
- Diagnosed Paula AI latency bottlenecks and proposed improvements using prompt caching, streaming response optimisation, and cleaner static/dynamic context separation.
- Collaborated on architecture for a self-improving AI cost-optimisation system designed to support product and investor-facing milestones.
🌍 Open Source - AI Infrastructure & Developer Tools Contributor
Contributed across Digital Crew Technologies, Tracer Cloud / OpenSRE, aden-hive, Lamatic, and independent MCP projects.
Highlights:
- OpenSRE: Slack/Discord delivery hardening, token redaction, non-JSON error handling, tests.
- aden-hive: improved onboarding documentation for uv workspace setup.
- Lamatic: contributed AgentKit configuration and metadata for automation workflows.
- Built independent MCP/devtool projects: GitHub PR Context MCP, Quant Brain MCP, hstack, killport.
📬 NSE News-to-Telegram Automation - Client Project
Built an automated financial news intelligence pipeline for real-time NSE market updates via Telegram.
Highlights:
- Web scraping infrastructure
- Financial news ingestion
- NLP-based filtering
- Telegram alert delivery
- Retry logic, monitoring, and production-grade error handling
| Attribute | Level | Evidence |
|---|---|---|
| AI Agent Infra | █████████░ 90% | MCP, routing, telemetry, agent workflows |
| Backend Engineering | ████████░░ 80% | FastAPI, Next.js, Supabase, service APIs |
| DevTools | ████████░░ 80% | hstack, killport, PR Context MCP |
| Production Debugging | ███████░░░ 70% | Auth, diagnostics, fallbacks, deployments |
| Open Source | ███████░░░ 70% | OpenSRE, aden-hive, Lamatic |
| Product Thinking | ████████░░ 80% | Max, Claire, Orchestrator, integrations, agent UX |
| Item | Power |
|---|---|
| MCP Servers | Give agents structured tools |
| Repo Memory | Makes agents remember past PRs and review patterns |
| Telemetry | Turns agent behaviour into learning data |
| Model Routing | Reduces cost without blindly sacrificing quality |
| Agent Workflows | Connects AI agents to real product operations |
| Prompt Caching | Reduces repeated context rebuilds |
| Streaming UX | Makes AI products feel faster |
| Backend Fallbacks | Keeps product flows alive when infra fails |
| Open Source | Proves public contribution and learning speed |
organizations/
├── Digital Crew Technologies → Max, Claire, Orchestrator, MCP, integrations
├── Tracer Cloud / OpenSRE → AI SRE reliability hardening
├── aden-hive → developer onboarding docs
├── Lamatic → AgentKit automation contribution
└── paarths-collab → MCP servers, agent tools, devtools
|
|
I post short notes on:
AI agents → what works, what breaks
MCP → servers, tools, workflows
backend infra → production lessons
AI workflows → agents, tools, integrations
startups → building in public
open source → contributions and experiments
If any of these tools made you curious, help me level them up:
| Repo | Mission | Action |
|---|---|---|
| GitHub PR Context MCP | Give coding agents repo memory | ⭐ Star |
| hstack | Deploy and wire Hermes agents | ⭐ Star |
| Quant Brain MCP | Finance intelligence over MCP | ⭐ Star |
| killport | Make CLI tools agent-readable | ⭐ Star |
🧬 why I build agents
I don't think the future is just chatbots.
I think the real opportunity is:
agents
+ tools
+ memory
+ workflows
+ permissions
+ telemetry
+ observability
+ production feedback loops
That is what I'm learning and building toward.If this made you curious, follow my build log.
I'm open to:
- AI agent infrastructure experiments
- MCP server ideas
- developer tools
- backend automation
- startup prototypes
- open-source AI infrastructure
- research-heavy AI products
Best first message:
Hey Paarth, I found your Agent Lab.
I'm building ______.
I think we can collaborate on ______.
|
killport Docker-native port killer with JSON output for humans, agents, and CI workflows. |
quant-brain-mcp MCP server for finance analysis, backtesting, portfolio intelligence, and chart generation. |
|
github-pr-context-mcp Repo memory for AI coding agents using PR history, review threads, and team patterns. |
hstack Portable skill catalog for AI coding agents, Hermes deployments, and integration workflows. |
built in public · shipped at odd hours · learning fast