ChaitanyaVootla

Staff engineer at Faeth Therapeutics. Mostly working on agents these days, with a day job in clinical-trial software.

Start a conversation View active project

scroll

10+Years writing software

AgentsWhere my time goes

Clinical trialsDay job

Product systemsWhere I’m strongest

What I work on

Three things, in this order.

Agent harnesses are where I spend my evenings and a lot of my Faeth time. Clinical-trial apps are the day job. Underneath both, ten years of normal full-stack product work.

Agent harnesses

Where I spend most of my time. Task contracts, tool boundaries, traces, evals — the scaffolding that turns a demo into something you can actually rely on. LangGraph is where I work.

Clinical trials

Day job at Faeth. Trial-related apps where getting it wrong has real consequences, so the work skews toward correctness, audit trails, and not breaking things in prod.

Full-stack

Ten years of product work before this — frontends, services, the cloud bits, the database underneath. Useful baseline for the agent stuff.

Timeline

Faeth Therapeutics

Staff Engineer

Current

Clinical-trial related applications and the internal tools around them. Healthcare-grade reliability, so a lot of the work is making sure the boring parts are actually boring.

JFrog

Senior Software Engineer

Apr 2019 — 2024

Artifactory and the CI/CD product line. Big Vue apps on the front, services and cloud orchestration on the back.

Shippable

Software Engineer

Jun 2016 — Apr 2019

Built parts of a CI/CD product at startup speed. Got acquired by JFrog mid-way through, which is how I ended up there.

NIT Agartala

B.Tech, CSE

2012 — 2016

Where I picked up the fundamentals before going full-time on product work.

Agents · what I’m on now

This is where most of my time goes.

Building the scaffolding around agents — task contracts, tool boundaries, traces, evals, the feedback loop. LangGraph is the framework I work in. It feels a lot like picking up React in 2017 — the obvious choice that just hasn’t been obvious for very long yet.

The work is the scaffoldingThe interesting bit isn't the model call. It's the contracts, the tools, the traces, and the evals around it.

Read the runs, not the demoLooks-good-once doesn't ship. If I can't see the runs over time, I can't tell you anything.

Domain firstAgents work best when they sit on top of a real model of the domain. The LLM is the thin layer.

Same boring disciplineObservability, rollbacks, audit trails. The stuff that turns 'cool demo' into 'in production'.

What I’m building on the side

TheMovieBrowser — the one I actually use every week.

Started as something to scratch my own itch around finding things to watch. Now it has a real backend, real recs, and a Chrome extension. Below is the live site.

Still adding things

Side project

01TheMovieBrowser.com

My evenings project. Movie and TV discovery with ratings from multiple sources, watch links, watchlists, custom filters, and recs backed by embeddings + vector search.

Ratings and watch linksCustom discovery filtersWatchlist and episode trackingOpenAI embeddings + Chroma recommendations

NuxtTypeScriptNodeMongoDBAWS LambdaOpenAIChroma

Visit site GitHub

Live TheMovieBrowser.com

Open site

The stack