Week 1 at devx
I'm an AI engineer who likes shipping models to the places they're hardest to run: phones. I joined devx to trade a narrow set of problems I understood for a much wider set I don't understand yet.
Who I am
I'm Gautam Rana. Most people call me RDx(Rana D' Extreme). I'm an AI Engineer in Nitin's pod (AI), and I joined on 1 September 2026.
I have a Bachleros in Computer Science. I spent my first year in industry at Propelius Technologies, six months as an intern and six months full time. By the end of that year I was leading the Applied AI works of the company as a go to 'AI' guy.
Outside work, I write articles, watch a lot of anime, and go looking for forests whenever I can. The long-term plan, if everything goes well, is to settle in Darjeeling.
The path that led here
Three things I've built
Wallee runs its model on the phone, not in the cloud. It's a Flutter app on the Play Store with more than 1,500 users, and still growing. Ingestion uses one-shot classification with CLIP models in Python with an ETL pipeline combined with the upscaler embedded. The model runs on-device through Rust, bridged into Flutter with flutter_rust_bridge. I'm now building a recommendation model for it. Wallee on the Play Store (download it i need more revenue by end of this year currently at $50).
At Propelius I built e-commerce search and the automations that feed it. That covered the search engine for harrir.com and fail-safe product-enrichment pipelines, all running in containers on Amazon ECS.
I also built multimodal voice agents on LiveKit. On the side, I'm writing a LiveKit agent SDK in Go. Goroutines make it cheap to run many agent processes at once and keep the I/O path fast.
How search pulled me into fine-tuning
Search work showed me how far an off-the-shelf model can be from a good ranking. So I implemented Amazon Science's paper on graph-based multilingual product retrieval in e-commerce search: a graph convolutional network, fine-tuned with an InfoNCE contrastive loss. That project is why I care about how models are trained, not only how they're called.
Why devx
I joined devx to convert low entropy into high entropy by creating knowledge pool on my brain basically that's my insta bio! LOL.
At Propelius I knew the shape of every problem that came my way. At devx the problems are wider, messier, and closer to enterprise customers. That's the growth I came for.
The first three weeks
Week one was chaos. Almost everything was new: Gemini Enterprise certifications, knowledge transfer on the CIMB engagement, working through the use cases, agent governance, and the agentic setup itself. I started with no picture of Gemini Enterprise, ADK deployment strategies, connectors, or how a private VPC connection works from the agent gateway. Those pieces are now fitting together.
Week two was basically digesting the CIMB Knowledge transfer
Week three was FDE training. We finished the forward-deployed engineering training as a cohort.
Week four was my first real code. I set up the Gemini Enterprise–Microsoft plugin and made the code changes it needed.
What I want to learn at devx
Core applied AI/ML and CS at scale.
Come to me for:
- On-device inference: running models on mobile through Rust and Flutter
- Real-time voice agents on LiveKit, and agent runtimes in Go
I want to learn from you: enterprise agent deployment on Google Cloud and AWS, governance, and how devx runs delivery with customers.
just find me as RDx.