AWS Builder Center
Everything Everywhere All at ₹500: Building SecondAct for Broke Artists

Everything Everywhere All at ₹500: Building SecondAct for Broke Artists

We make films in college. Our ideas need a production studio; our budget needs emotional support. So we built SecondAct with Gemini, FalkorDB and AWS to turn ambitious creative briefs into practical plans using available resources. Here’s how graph relationships, AI agents and serverless infrastructure help keep the idea alive when the sewing machine is booked.






Chapter 1:

Written by Christopher Nolan. Funded by the group chat.


We make films in college.
Which means we have experienced a very specific kind of heartbreak: coming up with an idea that looks incredible inside your head, then introducing it to your available resources.
The idea has atmosphere. Production design. Beautiful lighting. A shot that absolutely belongs in the trailer.
The resources have a phone, borrowed equipment and several people saying, “Wait, I think I know someone.”
We have faced this several times. The idea is next level. The resources are on another level entirely. Usually a lower one.
Then the negotiations begin.
Can we borrow that camera? Is it available on the day we shoot? Does it come with a lens? Can these curtains become a backdrop? Can we alter that costume, or does “borrow” mean return it with every molecule intact?
Somewhere between writing the scene and finding the equipment, you accidentally become a procurement department.
That experience became the starting point for SecondAct.



Chapter 2:

The villain is resource coordination


For student artists, having an idea and being able to execute it are two separate projects.
Filmmakers need cameras, props, costumes and lighting. Theatre teams need sets and materials. Photographers need backdrops and equipment. Fashion designers need fabric, tools and ways to assemble things without financially destabilizing themselves.
Often, useful resources already exist nearby.
Someone has the fabric. A club owns the equipment. Another team has leftover materials.
But finding out what exists is only the beginning.
You also need to know who owns it, when it is available, what it costs, what else it needs and what you are allowed to do with it.
A camera marked “available” is less exciting when its required lens is booked.
A free piece of fabric is less useful when your design requires cutting it and its owner would strongly prefer that you did not.
These details determine whether an idea survives.
SecondAct helps connect an artist’s creative brief to those practical constraints—and find another route when the first one is blocked.
The name fits twice: resources get another use, and ideas get another chance.



Chapter 3:

Meet SecondAct: ambition, with an equipment check

Landing Page of our website

SecondAct supports filmmaking, theatre, photography and fashion design.
You describe:
  • What you want to create.
  • What resources you already have.
  • Your budget.
  • Your project dates.
The app uses Gemini and a live FalkorDB graph to produce a plan with production steps, resource choices, checked borrowing costs, assumptions and unresolved needs.
You can also search the inventory directly without generating an AI plan.
“Use a sewing machine” is a suggestion.
“The sewing machine is booked during your dates, but a hand-sewing method has available supplies within your budget” is something you can act on.
That difference is most of the project.



Chapter 4:

The chatbot cannot manifest a camera


A language model can confidently suggest equipment that you do not own, cannot borrow or cannot afford.
It can write a beautiful production plan in which every practical problem has been solved by assuming the practical problem does not exist.
Excellent energy. Difficult shoot.
SecondAct gives the model tools to inspect inventory data before finishing its plan.
The responsibilities are straightforward:
Component
It's Job
Gemini
Explore creative approaches and choose tools
FalkorDB
Supply connected resource facts
Backend validation
Check the final selections
The artist
Decide whether the result serves the idea
The model handles open-ended creative work. The application checks the parts that can be checked.
Creativity remains flexible. Whether a lens is booked does not.



Chapter 5:

Everything is connected. Especially the equipment you forgot.

Actual Graph I Designed

A flat inventory can tell me that a camera, a lens and some curtains exist.
The relationships tell me whether I can make something with them.
SecondAct stores those connections in FalkorDB:
Club → OWNS → Camera
Camera → SUITABLE_FOR → Filming
Camera → REQUIRES → Lens
Booking → RESERVES → Lens
Follow those relationships and the catch becomes visible: the camera body may be free while the kit is unavailable.
Alternatives follow another path:
Blue fabric → SUITABLE_FOR → Backdrop
Teal curtains → SUITABLE_FOR → Backdrop
That shared purpose gives the app a reason to consider curtains as a replacement for backdrop fabric. It then checks their availability, quantity and cost.
This matters for artists because an object’s official name is often less interesting than what we can do with it.
A curtain can have a surprisingly active career in student cinema.
The current demo contains 35 fictional resources, along with clubs, materials, bookings, equipment dependencies and making techniques. The graph snapshot used for our visualization contained 99 nodes and 218 relationships.
These are sample records demonstrating the system, rather than a real campus inventory.

The graph also understands how something gets made

Finding material is only half the problem. You also need a workable method.
SecondAct connects techniques to their mandatory needs and the resources that can satisfy them:
Discipline → ENABLES → Technique
Technique → NEEDS → Need
Need → ACCEPTS → Resource
For fashion, this lets the app explore methods such as machine appliqué, hand appliqué and removable draping.
Here is a query from the project that exposes those connections in FalkorDB Browser:
MATCH (d:Discipline {key:'fashion'})
-[a:ENABLES]->(t:Technique)
-[b:NEEDS]->(n:Need)
-[c:ACCEPTS]->(r:Resource)
RETURN d,a,t,b,n,c,r
In plain English:
Show the fashion techniques, everything each technique needs, and the resources that could meet those needs.
Every mandatory need must be covered. Connected alternatives are choices; the artist does not have to borrow every item.
Resources also have explicit  canCut  and  canDye  permissions. A method requiring cut fabric cannot use borrowed fabric without cutting permission.
“Available” and “allowed to become seventeen smaller pieces” are separate properties.
That distinction felt necessary.



Chapter 6:

Hosting it on AWS without giving the infrastructure a bigger budget than the film.


The deployment uses a small hybrid architecture:
Browser
│
▼
Cloudflare Pages
│ /api/* through a Pages worker
▼
AWS Lambda — Mumbai
├── Gemini: creative planning
├── FalkorDB: connected inventory
├── DynamoDB: usage counters
└── CloudWatch Logs: operational logs
Cloudflare Pages serves the HTML, CSS and JavaScript frontend. Approved API requests pass through its worker to an AWS Lambda Function URL.
The backend runs in Mumbai ( ap-south-1 ), using Node.js 22 on ARM64 with 512 MB of memory. The Lambda adapter reuses the graph-search and planning modules used by the local server.
There is no always-on application server, VPC, NAT gateway, load balancer or API Gateway in this deployment. Gemini and FalkorDB remain external services, with credentials kept in private Lambda environment settings.
To keep the public demo manageable, DynamoDB tracks a shared limit of 30 AI planning attempts per UTC day, plus four attempts per IP per minute.
An atomic conditional update checks and increments each counter together. Otherwise, simultaneous requests could all decide they had claimed the final slot. Very democratic. Very expensive.
IP addresses are hashed before storage, and counters have expiry settings. Failed planning attempts count too; manual inventory search remains available after the daily AI limit is reached.
These controls limit application usage. They are not a hard spending cap, and actual costs depend on traffic, account terms and external-provider quotas.
CloudWatch Logs have seven-day retention. Application logging excludes briefs, model responses and secret values.
The aim was a reachable demo with a small operational footprint—and fewer opportunities for a student project to acquire a finance department.




Chapter 7:

Give the idea a second act.


SecondAct is a prototype. The inventory is fictional, profiles and briefs stay in the browser, and checking availability does not reserve anything. Production authentication, real borrowing transactions and persistent agent memory are still future work.
Connecting real inventory and reliable reservations would be the next step. Three film crews arriving for the same camera is a short film nobody volunteered to make.
Building this taught me that useful AI needs constraints it has to respect.
The model suggests an approach. The graph reveals dependencies. Backend validation checks stock and budget. The artist decides whether the result serves the idea.
We started SecondAct because we know what it feels like to have an ambitious college film idea and spend the next several days negotiating with reality.
Artists already improvise. Constantly.
A useful tool should help us see more possibilities in the resources around us—and give us enough evidence to start making something.
Because sometimes your cinematic universe really does begin with ₹500 and a curtain.



Before the credits:

we actually make movies.

Our Youtube Channel

The college filmmaking chaos mentioned above has a YouTube channel.
That’s us. Making films, figuring things out and occasionally asking a curtain to deliver a career-defining performance.
If you want to see the filmmaking behind the frustration that inspired SecondAct, come check out our work.
Our production budget would appreciate your emotional support.


Try it:  SecondAct  Explore the code:  GitHub  Go deeper:  Graph model  ·  Backend workflow  ·  AWS deployment 
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article