Sattyam Jain

Tech Lead · Production Agentic Systems · AI Platform Architecture

Ahmedabad, India · Daily US Pacific overlap · Open to relocation and visa sponsorship

sattyamjain96@gmail.com · sattyamjjain.in · github.com/sattyamjjain · linkedin.com/in/sattyamjain · x.com/sattyamjjain · medium.com/@sattyamjain96 · theproductionagent.substack.com

Summary

Tech Lead at Attri.ai with delivery oversight across 12+ active enterprise GenAI engagements in regulated US verticals (Insurance, Healthcare, Legal, EdTech, Construction, Geospatial). Architect of Agentify - Attri's flagship multi-agent platform with 15+ production agents - and the multi-cloud on-prem deployment mode that unlocked the regulated-vertical customer book. Created pyAGI, a Python package acquired by Kyle Morris and Jeffrey for $10,000. Coined the Capability Lease security pattern (short-lived, scope-narrowed agent credentials). Ship and maintain the agent-security OSS layer in active production use: agent-audit-kit (OWASP Agentic Top-10 SAST), agent-airlock (tool-call contract layer), and Provael (physical-AI red-team harness). Author of The Production Agent, serialized 2026.

Experience

Tech Lead - promoted Oct 2025 from Senior Software Engineer

Dec 2024 - Present

Attri.ai · Ahmedabad, India · Remote with daily US Pacific overlap

Architected Agentify, Attri's flagship production multi-agent platform, and lead the core team I hired for it: 15+ specialized agents (Orchestrator, PRD, Architect, Designer, Coder, Diff Analyzer, Discovery) coordinated by a central state-machine Orchestrator with full replay, audit, and human-in-the-loop checkpoints.
Built the multi-provider LLM routing layer across Anthropic Claude (Opus / Sonnet / Haiku), OpenAI GPT (incl. Pro), Azure OpenAI, and Google Gemini, and engineered the Cascading Router cost strategy (simple queries to small models, complex tool-use to Claude Opus).
Led E2B micro-VM sandboxing for agent code execution (built with the Agentify team) and MCP-style tool integration across 10+ services with standardized auth, contracts, and result normalization.
Designed the multi-tenant SaaS architecture (workspace segmentation, RBAC, Stripe billing, usage-based pricing, zero-downtime blue-green deployments).
Architected the multi-cloud on-prem fork (Terraform-per-environment, adapter pattern over identity, object storage, and datastore layers, jsonb metadata for client schema variance): one codebase across N client clouds, unblocking the regulated-vertical deployment mode.
Customer-facing technical lead with delivery oversight across 12+ enterprise engagements in six regulated US verticals (Insurance, Healthcare, Legal, EdTech, Construction, Geospatial), including a greenfield healthcare AI product taken from empty repo to v1 in 6 weeks with a pre-launch Infrastructure + Compliance Audit.
Coined the Capability Lease security pattern (short-lived, scope-narrowed, revocable agent credentials), adopted in agent-airlock and mnemo.
Built attri-dev-kit, the org-wide CI/CD platform rolling out across 208 internal repos, with AI-specific guardrails (hallucinated-import detection, AI-author signature escalation, unjustified-lint-disable hard-blocks, test-delta gate).
Built eval harnesses and OpenTelemetry → Datadog / Sentry / PostHog observability for agent reliability, tool-call accuracy, and latency across the 9 production AI projects where I am listed Tech Owner.
Allocation planning for 25+ engineers across 16 customer projects via a custom Resource Allocation Planner (delivery oversight, not direct reports), contributed performance reviews across a 22-person engineering org, and authored the org-wide Engineering Process Alignment doc adopted across every active project.

Software Engineer III

Dec 2023 - Nov 2024

Zenarate · Gurugram, India · Hybrid

Built conversational features for the contact-center training simulation platform on Azure Conversational Language Understanding (CLU) and LUIS: intent and entity handling for simulated customer conversations.
Optimized the AWS transcription pipeline for performance and cost.
Re-architected the simulation backend with health-gated rollouts and per-service circuit breakers for faster, safer deploys.
Designed REST + WebSocket APIs (FastAPI) for the real-time simulation layer; managed AWS infrastructure-as-code with high-availability rollouts.
Drove architecture decisions on service boundaries, session-tracking data models, and AI-assisted feature integration that did not compromise platform reliability.

Software Engineer Team Lead - promoted May 2023 from Software Developer

Dec 2020 - Nov 2023

MyShubhLife (FinTech) · Bengaluru, India

Built post-disbursal loan workflows in Kautilya, MyShubhLife's loan management system (disbursements, write-offs, waivers, refunds), working directly with the AVP and the finance and operations teams.
Proposed Heimdall, an internal mini ERP for operating Kautilya's loan workflows, to the AVP and CTO, then built it with the team.
Designed and shipped backend services for loan processing, customer onboarding, and financial analytics for the consumer-lending product (Python + Django/DRF + Celery + Postgres + Redis).
Ran OCR and NER proofs of concept for KYC document extraction (PAN, Aadhaar) and evaluated lower-cost extraction approaches with the data science team; picked HyperVerge on cost and results.
Designed APIs + architecture for loan lifecycle, credit-scoring integrations, and reporting; tuned database performance + monitoring for sustained throughput.
Led a team of four developers and two testers; co-authored pyluca - open-source double-entry accounting library (60+ ★); ran code reviews, facilitated design discussions, and bridged technical + non-technical stakeholders.

Projects

Selected open source and side projects

agent-airlock Type-checker and contract layer for AI agent tool calls, in-process: strict argument validation, ghost-argument stripping, self-healing retries, plus optional PII masking, RBAC, and sandboxed execution. Enforces Capability Lease policies in-process; v0.10.28 ships 7,999 tests at 90.33% coverage.
agent-audit-kit Static-analysis scanner for AI agents with full OWASP Agentic Top-10 and MCP Top-10 coverage (10/10 each) and SARIF output. 437 rules in total with mappings to 14 compliance frameworks (v0.6.29); Marketplace-listed, Sigstore-signed, with a public CVE-to-rule ledger of measured latency.
MannSetu Voice-first AI mental wellness companion for India in Hindi and English, with empathetic multi-turn dialogue and crisis-aware routing. Solo-built and live with early users; a wellness companion, not a substitute for professional care.
Provael Open-source red-team harness that attacks vision-language-action (VLA) robot policies in simulation, scoring attack-success with 95% Wilson intervals. Author of the Embodied AI Security Top 10; v0.46.0 registers 39 attacks over 17 families against 8 policy backends (one measured on a real SmolVLA checkpoint).
Why Can't We Have An Agent For This? Solo-built consumer AI tool that turns any everyday problem into an agent-feasibility analysis: verdict tiers, competitor research, and kill predictions. Launched Holi 2026; Hacker News was its #1 referrer in the first week.
255 (get255.com) Just-for-fun writing experiment on Jev, TypeSafe AI's decision model, via OpenRouter: one text, 255 AI impressions across voice, energy, craft, intent, and wildcards, with a change-one-sentence compare. Solo-built; live since September 2026.
VAJRA Physical AI side project, in progress: an autonomous find-and-fetch robot on Jetson Orin Nano with ROS 2, NanoOWL open-vocabulary detection, and behavior-tree planning. Applies Capability Lease governance instincts to physical actions.
pyAGI Python package for agentic loops (planning, memory, and tool use), published on PyPI in 2023 as one of the first open-source agent runtimes there. Acquired by Kyle Morris and Jeffrey for $10,000.

Skills

Languages: Python, TypeScript, JavaScript, Rust, SQL, Bash, Go (SDK-level)
LLMs and agents: Claude (Opus / Sonnet / Haiku), OpenAI GPT (incl. Pro), Google Gemini · LangGraph · Model Context Protocol (MCP) · Anthropic SDK · OpenAI SDK · RAG · multi-agent orchestration · custom eval harnesses · LLM-as-judge · AgentOps · cascading-router cost optimization
Agent security: OWASP Agentic Top-10, MCP Top-10, capability-scoped credentials (Capability Lease), prompt-injection defense, agent supply-chain auditing, CVE disclosure, SARIF-based SAST for agents; embodied-AI / VLA red-teaming (Provael), Embodied AI Security Top 10
Backend and data: FastAPI, Django/DRF, PostgreSQL, pgvector, Redis, Celery, DuckDB, USearch, Tantivy, vLLM (inference serving)
Cloud, platform, observability: AWS and Azure (primary), GCP, Docker, Terraform, E2B microVMs, GitHub Actions, OpenTelemetry, Datadog, SigNoz, Sentry
Compliance and regulated deployment: EU AI Act, SOC 2, HIPAA, NIST AI RMF, India DPDP, ISO 42001, ABA Model Rule 1.6 + Formal Opinion 512 (legal AI), on-prem + air-gapped deployment patterns
Physical AI (side projects): ROS 2 Humble, NVIDIA Jetson Orin Nano, NanoOWL (TensorRT-optimized OWL-ViT), CUDA, behavior trees; VLA policies (SmolVLA, OpenVLA, GR00T), LIBERO / robosuite / MuJoCo simulation, red-team evaluation harnesses

Education

Master of Computer Applications (MCA) · Madhav Institute of Technology and Science (MITS), Gwalior (affiliated with RGPV) · 2017 - 2020 · CGPA 8.87/10 · First Division with Honours
B.Sc. (Mathematics, Physics and Chemistry) · Bundelkhand University, Jhansi · 2014 - 2017

Talks and Writing

What Should You Measure Before Shipping an Adapted Policy? Attack Success Against a Benign Control Arm · NeurIPS 2026 RoboPAD workshop - Accepted as a poster (non-archival), Paris, December 2026.
Black Hat Europe 2026 Arsenal: provael, Reproducible Red-Team Assurance for VLA Robot Policies · github.com/provael/provael - Accepted tool demo, London, 9 December 2026.
The Capability Lease · sattyamjjain.in/writing/capability-lease - Manifesto + wire format for the pattern; adopted in agent-airlock and mnemo.
The Production Agent (book, serialized in public) · sattyamjjain.in/book - 5 planned chapters (2 drafted, 3 outlined) on orchestration, MCP security, supply-chain audit, agent-on-agent commerce.
The Embodied AI Security Top 10 (EAI01-EAI10) · provael.com - Independent risk taxonomy for vision-language-action robot policies (CC BY-SA 4.0, not affiliated with OWASP), authored as part of Provael.

Speaker availability: agentic AI, LLM evals, agent security, MCP governance. Request a slot at sattyamjain96@gmail.com.