Machine Learning

Constructing a Private API for Your Information Initiatives with FastAPI

April 22, 2025
roosho

have you ever had a messy Jupyter Pocket book stuffed with copy-pasted code simply to re-use some knowledge wrangling logic? Whether or not you do it for ardour or for work, when you code so much, you then’ve in all probability answered one thing like “approach too many”.

You’re not alone.

Possibly you tried to share knowledge with colleagues or plugging your newest ML mannequin right into a slick dashboard, however sending CSVs or rebuilding the dashboard from scratch doesn’t really feel right.

Right here’s immediately’s repair (and subject): construct your self a private API.
On this put up, I’ll present you how one can arrange a light-weight, highly effective FastAPI service to reveal your datasets or fashions and lastly give your knowledge tasks the modularity they deserve.

Whether or not you’re a solo Information Science fanatic, a pupil with facet tasks, or a seasoned ML engineer, that is for you.

And no, I’m not being paid to advertise this service. It’d be good, however the actuality is way from that. I simply occur to get pleasure from utilizing it and I assumed it was value being shared.

Let’s assessment immediately’s desk of contents:

  1. What’s a private API? (And why do you have to care?)
  2. Some use circumstances
  3. Setting it up with Fastapi
  4. Conclusion

What Is a Private API? (And Why Ought to You Care?)

99% of individuals studying this can already be aware of the API idea. However for that 1%, right here’s a short intro that will probably be complemented with code within the subsequent sections:

An API (Software Programming Interface) is a algorithm and instruments that enables completely different software program purposes to speak with one another. It defines what you possibly can ask a program to do, comparable to “give me the climate forecast” or “ship a message.” And that program handles the request behind the scenes and returns the outcome.

So, what’s a private API? It’s basically a small internet service that exposes your knowledge or logic in a structured, reusable approach. Consider it like a mini app that responds to HTTP requests with JSON variations of your knowledge.

Why would that be a good suggestion? For my part, it has completely different benefits:

  • As already talked about, reusability. We are able to use it from our Notebooks, dashboards or scripts with out having to rewrite the identical code a number of instances.
  • Collaboration: your teammates can simply entry your knowledge via the API endpoints without having to duplicate your code or obtain the identical datasets of their machines.
  • Portability: You possibly can deploy it anyplace—domestically, on the cloud, in a container, and even on a Raspberry Pi.
  • Testing: Want to check a brand new function or mannequin replace? Push it to your API and immediately check throughout all shoppers (notebooks, apps, dashboards).
  • Encapsulation and Versioning: You possibly can model your logic (v1, v2, and so forth.) and separate uncooked knowledge from processed logic cleanly. That’s an enormous plus for maintainability.

And FastAPI is ideal for this. However let’s see some actual use circumstances the place anybody such as you and me would profit from a private API.

Some Use Circumstances

Whether or not you’re a knowledge scientist, analyst, ML engineer, or simply constructing cool stuff on weekends, a private API can change into your secret productiveness weapon. Listed here are three examples:

  • Mannequin-as-a-service (MASS): practice an ML mannequin domestically and expose it to your public via an endpoint like /predict. And choices from listed here are infinite: fast prototyping, integrating it on a frontend…
  • Dashboard-ready knowledge: Serve preprocessed, clear, and filtered datasets to BI instruments or customized dashboards. You possibly can centralize logic in your API, so the dashboard stays light-weight and doesn’t re-implement filtering or aggregation.
  • Reusable knowledge entry layer: When engaged on a mission that accommodates a number of Notebooks, has it ever occurred to you that the primary cells on all of them comprise at all times the identical code? Nicely, what when you centralized all that code into your API and acquired it completed from a single request? Sure, you could possibly modularize it as nicely and name a operate to do the identical, however creating the API means that you can go one step additional, having the ability to use it simply from anyplace (not simply domestically).

I hope you get the purpose. Choices are infinite, identical to its usefulness.

However let’s get to the fascinating half: constructing the API.

Setting it up with FastAPI

As at all times, begin by organising the atmosphere together with your favourite env device (venv, pipenv…). Then, set up fastapi and uvicorn with pip set up fastapi uvicorn. Let’s perceive what they do:

  • FastAPI[1]: it’s the library that may enable us to develop the API, basically.
  • Uvicorn[2]: it’s what’s going to enable us to run the online server.

As soon as put in, we solely want one file. For simplicity, we’ll name it app.py.

Let’s now put some context into what we’ll do: Think about we’re constructing a sensible irrigation system for our vegetable backyard at dwelling. The irrigation system is kind of easy: we’ve a moisture sensor that reads the soil moisture with sure frequency, and we need to activate the system when it’s under 30%.

After all we need to automate it domestically, so when it hits the brink it begins dropping water. However we’re additionally curious about having the ability to entry the system remotely, possibly studying the present worth and even triggering the water pump if we need to. That’s when the private API can turn out to be useful.

Right here’s the fundamental code that may enable us to do exactly that (notice that I’m utilizing one other library, duckdb[3], as a result of that’s the place I might retailer the info — however you could possibly simply use sqlite3, pandas, or no matter you want):



import datetime

from fastapi import FastAPI, Question
import duckdb

app = FastAPI()
conn = duckdb.join("moisture_data.db")

@app.get("/last_moisture")
def get_last_moisture():
    question = "SELECT * FROM moisture_reads ORDER BY day DESC, time DESC LIMIT 1"
    return conn.execute(question).df().to_dict(orient="information")

@app.get("/moisture_reads/{day}")
def get_moisture_reads(day: datetime.date, time: datetime.time = Question(None)):
    question = "SELECT * FROM moisture_reads WHERE day = ?"
    args = [day]
    if time:
        question += " AND time = ?"
        args.append(time)
    
    return conn.execute(question, args).df().to_dict(orient="information")

@app.get("/trigger_irrigation")
def trigger_irrigation():
    # It is a placeholder for the precise irrigation set off logic
    # In a real-world state of affairs, you'll combine together with your irrigation system right here
    return {"message": "Irrigation triggered"}

Studying vertically, this code separates three fundamental blocks:

  1. Imports
  2. Organising the app object and the DB connection
  3. Creating the API endpoints

1 and a pair of are fairly easy, so we’ll deal with the third one. What I did right here was create 3 endpoints with their very own features:

  • /last_moisture exhibits the final sensor worth (the newest one).
  • /moisture_reads/{day} is helpful to see the sensor reads from a single day. For instance, if I wished to match moisture ranges in winter with those in summer time, I might examine what’s in /moisture_reads/2024-01-01 and observe the variations with /moisture_reads/2024-08-01.
    However I’ve additionally made it capable of learn GET parameters if I’m curious about checking a particular time. For instance: /moisture_reads/2024-01-01?time=10:00
  • /trigger_irrigation would do what the identify suggests.

So we’re solely lacking one half, beginning the server. See how easy it’s to run it domestically:

uvicorn app:app --reload

Now I might go to:

However it doesn’t finish right here. FastAPI gives one other endpoint which is present in http://localhost:8000/docs that exhibits autogenerated interactive documentation for our API. In our case:

It’s extraordinarily helpful when the API is collaborative, as a result of we don’t have to examine the code to have the ability to see all of the endpoints we’ve entry to!

And with only a few traces of code, only a few in truth, we’ve been capable of construct our private API. It will probably clearly get much more sophisticated (and possibly ought to) however that wasn’t immediately’s goal.

Conclusion

With only a few traces of Python and the facility of FastAPI, you’ve now seen how straightforward it’s to reveal your knowledge or logic via a private API. Whether or not you’re constructing a sensible irrigation system, exposing a machine studying mannequin, or simply bored with rewriting the identical wrangling logic throughout notebooks—this strategy brings modularity, collaboration, and scalability to your tasks.

And that is just the start. You would:

  • Add authentication and versioning
  • Deploy to the cloud or a Raspberry Pi
  • Chain it to a frontend or a Telegram bot
  • Flip your portfolio right into a residing, respiration mission hub

Should you’ve ever wished your knowledge work to really feel like an actual product—that is your gateway.

Let me know when you construct one thing cool with it. And even higher, ship me the URL to your /predict, /last_moisture, or no matter API you’ve made. I’d like to see what you give you.

Assets

[1] Ramírez, S. (2018). FastAPI (Model 0.109.2) [Computer software]. https://fastapi.tiangolo.com

[2] Encode. (2018). Uvicorn (Model 0.27.0) [Computer software]. https://www.uvicorn.org

[3] Mühleisen, H., Raasveldt, M., & DuckDB Contributors. (2019). DuckDB (Model 0.10.2) [Computer software]. https://duckdb.org