Staff Engineer, AI Platform (Agentic & Distributed Systems)

Mstack is an AI-native specialty chemicals company. Our R&D; platform, Chemstack, lets a chemist draw a target molecule and see ranked synthesis routes, with predicted yield, cost and impurities, before anything runs on the bench. From there it takes the molecule all the way to a kilo-scale procedure ready for manufacturing. We work with pharmaceutical, agrochemical and specialty-chemical companies.

The role:

Much of Chemstack's work is done by agents. They search literature and patents, run simulations, and narrow thousands of possible routes down to the one experiment worth running. We're hiring a Staff Engineer to own the architecture underneath them: the B2B platform, the security and tenancy model, and the agent runtime. The goal is a system that stays reliable, secure and affordable as we grow from a handful of pilot customers to many.

You'll lead the platform team of 5–8 engineers that does this work: setting technical direction, growing the engineers on it, and staying close enough to the code to write the hard parts yourself when it matters.

Why Mstack:

- Early, with real users. Chemstack is in pilots with five companies, and you'll hear directly from the chemists using it.
- Chemistry and software in one team. Founded by Shreyans Chopra in 2022, with scientific leadership from Prof. Gautam Desiraju (Ewald Prize) and R&D; and data-science leaders from SRF and Myntra.
- Funded and growing. A $40M Series A co-led by Lightspeed and Alpha Wave, 10x revenue growth, and operationally profitable.
- Decisions that last. The platform is live, and the architecture that takes it from pilots to scale is still being designed. Your choices will shape it for years.

What you'll do:

- Own the platform architecture. Design the B2B platform end to end: services, APIs, data model and the boundaries between them. Make sure it holds up as we grow from five customers to hundreds, and from a few route campaigns to thousands running at once across agents and GPU/HPC compute.
- Build for distributed scale. Set the standards for queues, event streams, scheduling, consistency and failure recovery. Simulation jobs that run for hours need to survive restarts and retries, and every expensive step should be resumable.
- Make multi-tenancy and security first-class. Customer molecules are confidential IP. You'll own tenant isolation, SSO and authorization, encryption and key management, secrets, audit logging,



and the controls enterprise security reviews ask for.
- Get the platform enterprise-ready. Design the deployment models large customers ask for: multi-tenant SaaS, single-tenant and VPC, and on-premises or air-gapped installs for customers who can't send molecules to the cloud. Own business continuity and disaster recovery: backups, multi-region failover, and RPO and RTO targets we can commit to in contracts.
- Build compliance into the platform. Work with security and legal on SOC 2 and ISO 27001, data residency, retention and deletion, and audit trails that hold up in customer due diligence. Pharma customers will ask about GxP and 21 CFR Part 11, so design workflows that are validated and traceable.
- Design the agent runtime. The tool layer and MCP integrations, an LLM gateway with routing, caching and cost controls, context and memory management, and guardrails against prompt injection and unsafe actions.
- Make AI behaviour observable and testable. Tracing that shows why an agent did what it did, versioning for prompts and models, and eval suites in CI that catch regressions before customers do.
- Lead the platform team. Set the technical direction and roadmap, run design reviews, hire, mentor and grow the engineers on the team, and work closely with product, chemistry and ML teams.

What we're looking for:

- 10+ years building and operating production software, including at least three owning distributed systems with live traffic, on-call and real postmortems.
- You've led a team of five or more engineers through a major platform build, setting direction and growing people while staying close to the code.
- You've designed a multi-tenant B2B SaaS platform for enterprise customers and can explain the tenancy model you chose, what it cost, and what you'd change.
- You've shipped at least one agentic or LLM-powered system to real users, and you know the AI layer well: tool calling, context engineering, RAG, evals, and the cost and latency trade-offs between models. You can tell us how it failed and what you changed.




- You've built security into a platform rather than bolting it on (authN/authZ, tenant isolation, key and secrets management, threat modelling), and you're comfortable walking an enterprise customer's security team through it.
- You're solid in Python or Go and have run production on at least one major cloud (AWS, Azure or GCP) with Kubernetes and infrastructure as code.
- You write clearly, and people come to you for judgment on hard calls.

Nice to have:

- Built or standardised MCP servers or similar tool protocols across teams.
- Built eval, tracing or LLM-gateway platforms that other teams relied on.
- Scheduled scientific or HPC workloads (DFT, molecular dynamics), or worked closely with scientists.
- Experience with GxP or 21 CFR Part 11, or validating software for other regulated industries.
- Shipped software that runs in customers' own environments (on-premises, private cloud or air-gapped), with Helm charts or operators, upgrades and support across many versions.
- Designed business continuity and disaster recovery for a production platform: backups and restores, multi-region failover, RPO and RTO targets, and DR drills that actually ran.
- Designed the controls for enterprise compliance such as SOC 2, ISO 27001, GDPR or India's DPDP Act.
- A background in chemistry, life sciences or another scientific field.

Your first six months:

- First 30 days. Learn the live platform alongside our chemists, pilot customers and your team, find where it will strain first, and write the target design for scaling tenancy, security and the agent runtime beyond the pilots.
- By three months. Your team has closed the highest-risk gaps from that design in production, such as tenant isolation, tracing across every agent run and evals in CI, without disrupting pilot customers.
- By six months. Under your lead, the platform is ready for broad enterprise rollout, has passed its first enterprise security reviews, and has proven its DR plan in a real failover drill. The team runs on the roadmap and design reviews you put in place.

How we hire:

1. Intro call (30 min) with the hiring manager.
2. System design (90 min): a multi-tenant agent platform for a chemistry use case. We focus on scale, isolation, security and failure modes.
3. Deep dive (60 min) on a system you built. We'll go as deep as you can take us.
4. Hiring manager round (45 min): scope, expectations for the first six months, and how you like to work.

📌 Staff Engineer (Bengaluru)
🏢 Mstack AI
📍 Bengaluru