AI FULLSTACK ENGINEER · AVAILABLE FOR WORK

I love chasing hard problems (the kind I'll bang my head against and lose sleep over), then shipping my way out of them. Most recently at Shoppin, where I built an engine that turns a garment photo into a factory-ready sewing pattern in minutes, work that used to take a master pattern-maker days, plus the AI systems around it: the feed, the try-on, the payments. Founding engineer at every startup before this one. I ship fast, I'm at my best debugging the thing nobody else can find, and I'm happiest watching a p95 graph fall.

// currently

~/now.md updated Jul 15, 2026
buildinga product of my own: self-healing agents + evals for multi-channel conversational AI
writingTessera, an agent-memory library, plus the evals to benchmark it against the top providers
readingeverything I can find on how GPU inference actually works
listeningdaily.dev + tech podcasts, keeping a finger on what's shipping
learningGPU inference, properly and in depth
Currently between roles and building, deep in agent memory and how GPU inference actually works, hunting for the next hard system to own.
SHIPS · LAST 90 DAYS
location New Delhi, India
local time 00:17 IST
status open to work

// the tape

3+
years shipping
3
startups · 2× founding eng
720+
github stars · Axon
14×
latency cut · /profile/me
250K+
items on the feed

// where I've been

// selected projects

// what I build with

languages PythonTypeScriptJavaScriptGoSQLC / C++
ai / ml RAGagent memoryagent harnessesevalsguardrails / OWASP-LLMLLM tracingmodel & tool routingcontext engineeringmulti-agent orchestrationMCPLangChainLangGraphLlamaIndexPyTorchHugging FaceLLM APIs (OpenAI · Anthropic · Gemini)
frontend ReactNext.jsReact NativeTailwindHTML / CSS
backend FastAPINode.jsExpressDjangogRPCCeleryRabbitMQPrismaSQLAlchemyWebSockets / SSE
data PostgreSQLRedisKafkaDebeziumApache PinotElasticsearchMongoDBMilvus / Zillizpgvector
infra / devops KubernetesDockerAWSAzureGCPCloudflare WorkersRayGitHub ActionsOpenTelemetryPrometheusGrafanaSentryPlaywrightLinux

// currently tinkering

001 self-healing agents my own product: self-healing agents + evals for multi-channel conversational AI [wip]
002 axon v2 a bigger, faster rewrite of the code-graph engine [wip]
003 ai-ugc-studio brand brief → finished, post-produced AI UGC video, end to end [shipping]

// two ways to read this

harshkedia.com ● YOU'RE HERE
The rendered build: polished, interactive, the full experience.
harsh.md
The same content, rendered as its markdown source: cleaner, quieter, on a memorable domain.
or just grab the résumé →
STILL HERE?

Say hi.

harshkedia717@gmail.com