0→1 agent infrastructure · 1→100 production hardening
Shreyaan Seth · Engineering Lead
I build agent runtimes and the backend around them: execution state, tools, memory, retrieval, evals, migrations, observability, and recovery when a provider fails.
Case studies
What broke, the options I rejected, what changed, and the result.
Agent infrastructure
A production agent runtime sent its execution state to a second LLM just to write the final answer. I wrote the RFC to fold synthesis into the primary runtime and replace agent-level debugging with step-level tracing.
Distributed systems
A deploy could delete a workflow in the middle of a two-day wait. I traced the problem to in-memory continuations and moved the runtime to persisted waits and duplicate-safe resume claims.
Multi-tenant security
An AI SQL agent queried shared multi-tenant tables. I designed and implemented tenant isolation with PostgreSQL Row-Level Security, from authenticated context through connection-pool safety.
Cloud infrastructure
A production database cost about $988 a month while averaging 0.8% CPU and barely using its 30,000 provisioned IOPS. I right-sized compute and storage performance, bringing the RDS bill down to about $320.
Runtime thinking
The model sits inside a larger system: state, retries, costs, tool side effects, and enough history to debug a run after the user closes the tab.
agent_runtime / run_01J7
LIVEDurable run state
The job has an identity, an owner, and a recoverable starting point before execution begins.
Also built
A few more systems I've worked on, and what I owned in each one.
Built the reusable evaluation system
Golden and adversarial datasets, deterministic tool mocks, retrieval fixtures, and structured or model-judged assertions for the places agent regressions usually hide.
Implemented the integration and scope layer
Durable retention and recall across tenant, user, external subject, agent, and knowledge-base boundaries, with rolling compaction for conversations that outgrow a model window.
Delivered the feature end to end
Kept business metrics deterministic while surrounding model output with provenance checks, critic loops, confidence signals, evidence inspection, and actionable enterprise UX.
Designed the staged migration
Separated raw provider capture from application transforms, then added incremental cursors, backfills, reconciliation tests, and explicit token and callback boundaries across Python and TypeScript.
Implemented and hardened critical paths
Built resilient browser recording with local session state, chunked upload queues, retries, and live transcription; hardened a telephony-to-realtime-AI bridge around server-side speech detection.
Operating principles
Logs, traces, costs, and run history are how a system explains itself when the demo ends.
Durable workflows get simpler when progress, failure, and ownership are visible by design.
Retries, cancellation, recovery, and isolation shape the experience as much as the happy path.
I like small boundaries, direct control flow, and infrastructure a tired engineer can reason about.
Experience
Engineering Lead · Backend & AI Infrastructure
Building major product surfaces from React and Next.js through agent runtimes, retrieval, real-time workflows, enterprise boundaries, and production infrastructure.
Software Engineer
Cut a critical API path from roughly seven seconds to 400 milliseconds, improved Next.js performance, and moved delivery to self-hosted infrastructure.
Software Developer
Built a consumer-safety browser extension that identifies suspicious websites and explains the warning instead of silently blocking the user.
Quick answers
Direct answers for founders, engineering leaders, recruiters, and the systems they use to find technical talent.
Currently open to the right problem
I'm interested in backend, AI infrastructure, platform, full-stack product, and early engineering roles where ownership is real and reliability matters.
shreyaans20@gmail.com