From Arm Pain to AI Gateway: Why I Chose Portkey for Managing Multiple AI Providers

Managing multiple LLMs meant juggling auth, errors, and APIs. Instead of managing APIs, I chose Portkey's AI gateway to handle the infra, so I could build Dictation Daddy!

From Arm Pain to AI Gateway: Why I Chose Portkey for Managing Multiple AI Providers


The Problem Started With My Arm

Last year, I couldn't type anymore.

Two years of building products had finally caught up with me. Every keystroke sent sharp pain through my arm. So like any developer would, I decided to build my way out of it.

I tried Mac's built-in dictation first. Honestly, it was pretty bad. But when I hooked up OpenAI's Whisper model, the transcription accuracy became incredible. I showed it to a few developer friends who were also typing all day. They loved it. That's how Dictation Daddy started. Not from some big market opportunity, but from actual physical pain.

Building Was Easy. Managing AI Providers Was a Nightmare

Here's something I learned the hard way: coding the product is actually the easy part.

I had a working app with Whisper for transcription. But transcription was just the first step - I needed post-processing to clean up the output, format it properly, and make it actually useful. This meant experimenting with different AI models. OpenAI for some tasks, Groq for others when I needed faster responses, Claude for more complex formatting.

Each provider had different APIs, different authentication methods, different error handling. I was basically building my own janky API gateway instead of focusing on my actual product. And the worst part? Every millisecond of latency mattered for a real-time dictation app. Users speak, and they expect to see text immediately. Any delay in transcription or post-processing killed the experience.

A Conversation That Changed My Approach

I was getting frustrated managing all these different AI providers, especially since I knew there were good solutions for Python but couldn't find anything solid for Node.js. So I messaged in The Generative AI Group asking how people handle this kind of thing in Node.

That's when Vrushank introduced me to Portkey.

He explained how it works as an API gateway that sits between your app and all your AI providers. One interface for everything. Built-in monitoring to see what's actually happening with your prompts. The ability to add evaluations to track output quality.

The Actual Integration

I want to share how simple the technical setup was:

First, I connected all my AI providers through Portkey's dashboard. OpenAI, Groq, Claude - all in one place. Then I replaced all my different API calls with Portkey's unified interface. Same code structure whether I'm calling Whisper for transcription or GPT-4 for post-processing.

The whole migration took maybe an hour. Not days of refactoring. An hour.

But here's what really blew me away: it didn't add any noticeable latency. For a real-time application like dictation, this was critical. When someone speaks, the audio needs to be transcribed immediately, then post-processed instantly. Portkey's AI gateway handled all this routing without adding delay.

The Game-Changing Features

Monitoring That Actually Helps
Before Portkey, I had no idea what was really happening with my prompts. Now I could see exactly which prompts were working, which were failing, and why. When users complained about formatting issues, I could trace back to the exact prompt and model combination that caused it.

Evaluations Built In
I added evals to automatically check the quality of my outputs. Is the transcription accurate? Is the formatting consistent? Are there any obvious errors? Instead of manually checking outputs or waiting for user complaints, I had automatic quality checks running on every response.

Experimenting Without Breaking Things
I needed to experiment with a lot of models for post-processing. Some were better at fixing grammar, others at maintaining the speaker's voice, some were faster but less accurate. With Portkey, everything used the same interface. I could swap models with a config change, no code rewrite needed.

When Support Actually Matters

Here's a story that shows what good support looks like:

I needed Portkey to support audio streams for real-time transcription. This wasn't a standard feature at the time. I reached out to their team explaining my use case.

Within 2 days, they had raised a PR adding audio support.

Two days. Not "we'll add it to our roadmap." Not "that's an enterprise feature." They just built it.

What Actually Happened

Within days of integrating Portkey, I was able to:

  • Reduce latency by finding the optimal model for each step of processing
  • Improve transcription quality by A/B testing different post-processing approaches
  • Actually understand why certain transcriptions were failing through proper observability
  • Add new AI features without worrying about provider lock-in

Each improvement was based on real data from the monitoring and evaluation system, not guesswork.

Looking Back

It's kind of funny how this all started. My arm hurt from typing too much, so I built a dictation app. Then I needed a way to manage multiple AI providers without drowning in complexity. Both problems had the same theme: removing unnecessary friction.

Dictation Daddy removes the friction of typing. Portkey removed the friction of managing AI providers. Sometimes the best solutions are the ones that just get out of your way and let you focus on what matters.

Building AI products should be about crafting great experiences, not managing infrastructure. That's the whole reason I went with Portkey, and honestly, it's one of the better technical decisions I've made.

These days, when friends ask me about managing multiple AI providers, I tell them to stop trying to build it themselves. Pick something that handles the complexity while maintaining performance. Worry about infrastructure optimization when you have enough users to make it matter.

For me, that meant going from arm pain to a fully monitored, multi-provider AI application that actually performs in real-time. Not bad for something that started as a personal problem.


Rahul is the creator of Dictation Daddy, an AI powered dictation tool that has very high accuracy compared to traditional tools like Dragon Dictation. It's used by professionals, doctors, lawyers etc to write 3 times faster.