Portfolio • Book a Call • LinkedIn • X • Email
I'm a GenAI Architect & Tech Lead at Attri.ai building production-grade AI platforms that ship real outcomes, not demos. I focus on agent orchestration, reliability, cost efficiency, observability, and agent security, for both software agents and physical AI.
Highlights
- Provael: red-team & assurance for physical AI (attacks VLA robot policies, reports attack-success rate with a 95% CI). I authored the Embodied AI Security Top 10. provael.com
- Agentify: Attri.ai's flagship multi-agent platform (15+ specialized agents); ~70% LLM cost cut, 99.9% uptime via a cascading router + E2B micro-VM sandboxing
- MannSetu: solo-built voice-first AI mental wellness platform for India (50+ active users · 10K+ conversations · 4.8/5)
- VAJRA: autonomous find-and-fetch robot (Jetson Orin Nano + ROS 2 + NanoOWL), building in public
- pyAGI: early open-source Python agent framework (2023), acquired 2025 by Kyle Morris (banana.dev co-founder) & Jeffrey
If you want a 10-second summary: I build agentic systems that work in production, software and physical.
| Domain | Details |
|---|---|
| Multi-Agent Systems | Role design, tool calling, memory, governance, deterministic workflows |
| GenAI Engineering | Cascading model routing, budget controls, evals, guardrails, RAG, micro-VM sandboxing |
| Agent Security | Static scanning, runtime firewalls, capability leases, OWASP Agentic Top-10, red-teaming |
| Cloud Architecture | AWS + Azure (primary), GCP, multi-tenant SaaS, IaC, queues, caching |
| Ops Mindset | Observability (OpenTelemetry, Datadog, Sentry), incident response, cost visibility, SLAs |
I'm building the open-source security infrastructure for AI agents: static analysis, runtime firewalls, and operational control planes, now extending to embodied, physical AI.
| Layer | Project | What It Does |
|---|---|---|
| Pre-deployment scanning | agent-audit-kit | Static security scanner: 77 rules, SARIF, OWASP 10/10, taint analysis |
| Runtime firewall | agent-airlock | Validates tool calls, strips ghost args, PII masking, RBAC, sandboxing |
| Operations control plane | ferrumdeck | Rust governance engine, budget enforcement, deny-by-default policies |
| Agent memory | mnemo | MCP-native memory database, semantic search, context windowing |
| Browser security | zerotrust | On-device AI threat detection, privacy-first, runs entirely in browser |
| Embodied / physical AI | Provael | Red-teams VLA robot policies; attack-success rate with 95% CI; I authored the Embodied AI Security Top 10 |
|
agent-audit-kit Security scanner for MCP-connected AI agent pipelines. 77 rules, 13 scanners, OWASP Agentic 10/10, GitHub Action, SARIF, taint analysis, tool poisoning detection. 
|
agent-airlock Security firewall for AI agents: validates tool calls, strips ghost arguments, enforces type safety, PII masking, RBAC, cost tracking & sandbox isolation. 
|
ferrumdeck Production-grade AgentOps control plane. Rust governance engine + Python LLM runtime + Next.js dashboard. Deny-by-default policies, budget enforcement & audit logging. 
|
|
mnemo MCP-native AI memory database written in Rust. Persistent conversational memory for AI agents with semantic search, context windowing & efficient vector storage. 
|
zerotrust On-device AI Chrome extension for real-time website security analysis. Privacy-first threat detection that runs entirely in the browser; no data leaves your machine. 
|
vajra Autonomous find-and-fetch robot. Jetson Orin Nano + ROS 2 + NanoOWL vision. Named after Indra's thunderbolt. Building in public. 
|
More Notable Projects
| Project | What it is | Links |
|---|---|---|
| pyAGI | Early open-source Python agent framework (2023); acquired 2025 by Kyle Morris & Jeffrey | Repo ·  |
| inksight | Forensic signature verification with explainable AI (pip install inksight) |
Repo ·  |
| AgentX | Unified agent framework playground (tools + eval + obs) | Repo ·  |
| PrimeAcrossAll | Same problem across many languages (Docker-based) | Repo ·  |
| Dynamic-ML-Orchestration | Serverless ML orchestration patterns | Repo ·  |
| PyVerseAI | Collection of AI + cloud projects | Repo ·  |
- Serialized book: The Production Agent, how to ship agentic AI to production without getting paged at 3am (5 chapters, 1-2 free)
- Medium: medium.com/@sattyamjain96
- Dev.to: dev.to/sattyamjjain
- Website: sattyamjjain.in
I'm open to:
- Advisory + consulting (agent architecture, security, cost optimization)
- Open-source collaborations (especially agent governance & physical AI)
- Speaking at conferences and meetups on production AI systems
