From Decomposition to Data: Building an AI-Assisted Forensic Tool for a Local Problem

What if a small colour-changing strip, photographed with an ordinary smartphone, could eventually provide a forensic investigator with useful information about the stage of decomposition?

DECOMPOSED REMAINS

       ↓

Biological changes occurring

       ↓

Microbial + environmental signals

       ↓

Conventional workflow

       ↓

Sample → Transport → Laboratory → Analysis → Interpretation

       ↓

        TIME

That question began with a problem I encountered not in the world of artificial intelligence, but in forensic science.

I work in forensic science, where some of the most important questions often have to be answered with incomplete information. When decomposed human remains are discovered, investigators may need to understand how far decomposition has progressed and what biological processes are occurring. Microorganisms are an important part of that process: bacterial and fungal communities change as decomposition progresses.

But there is a practical challenge.

Much of the information that could potentially be useful requires sampling, laboratory processing, microbiological analysis and interpretation. In a field investigation, that can mean waiting for information that may not be immediately available.

So I began asking a simple question:

Could we turn biological signals of decomposition into information that is faster, more accessible and potentially usable closer to the scene?

That question eventually led me from experimental decomposition research toward an AI prototype.

The problem starts with something we usually cannot see

Decomposition is not a single event. It is a dynamic biological process involving tissues, microorganisms, environmental conditions and chemical changes.

As decomposition progresses, microbial populations can change. Temperature, moisture, soil characteristics and other environmental factors can also influence those changes.

For a forensic scientist, this creates an interesting possibility.

If microbial patterns change systematically over time, perhaps those patterns could become another source of information for understanding decomposition.

The challenge is that biological data are rarely as simple as:

"One measurement equals one answer."

Instead, there may be dozens of interacting observations.

That is where artificial intelligence becomes interesting—not as a replacement for forensic science, but as a tool for finding patterns within complex data.

Conceptual illustration adapted from existing applications of AI and machine learning in biological analysis; it illustrates the analytical principle rather than the current prototype. https://www.preprints.org/manuscript/202304.0209

Starting with an experimental model

Before thinking about an AI application, I needed something much more fundamental:

data.

My existing research involves studying microbial changes during decomposition using an experimental animal model.

The purpose was not to claim that an animal decomposition model is equivalent to human decomposition. It isn't.

Instead, the model provides a controlled way to explore a scientific question:

Do measurable microbial changes occur at different stages of decomposition, and can those changes provide useful temporal patterns?

The preliminary observations were encouraging.

Across the experimental period, bacterial and fungal measurements did not remain static. Different microbial groups showed increases, peaks and declines at different points in the decomposition process.

That created something more interesting than a collection of laboratory measurements.

It created a pattern.

And that led to the next question:

Could an AI system learn that pattern?

From measurements to a microbial signature

The first version of the idea is deliberately modest.

Instead of asking an AI system to predict a precise postmortem interval, the prototype focuses on something more scientifically defensible:

decomposition-stage profiling.

The system can potentially consider multiple observations together, such as:

• bacterial measurements;
• fungal observations;
• temperature;
• moisture;
• pH;
• environmental conditions; and
• other experimental variables.

Rather than looking at each measurement independently, an AI-assisted model can examine their combined relationships.

The intended output is not:

"The person died 8.4 days ago."

That would be an unjustified claim at this stage.

Instead, the prototype aims toward something more cautious:

"This microbial and environmental pattern is most compatible with an early, active, advanced or late decomposition stage."

That distinction matters.

A research prototype should not be presented as a validated forensic instrument before it has undergone appropriate testing.

Why AI?

For me, the attraction of AI is not that it makes the project sound futuristic.

It is that decomposition produces multidimensional information.

A human researcher can examine individual variables, plot trends and identify obvious changes. But as the number of variables increases, the relationships become harder to interpret manually.

AI offers a way to explore those relationships systematically.

The goal is to use Google AI technologies to help transform experimental observations into an interpretable decomposition profile.

The broader workflow looks like this:

Experimental observations

↓

Microbial + environmental data

↓

AI-assisted pattern recognition

↓

Decomposition-stage profile

↓

Researcher-facing interpretation

The important part is the middle.

The AI is not being asked to replace the forensic scientist. It is being asked to help identify patterns within biological information that could otherwise be difficult to interpret quickly.

From numbers to colours

But there is another part of this idea that excites me even more.

What if the system did not always require a laboratory spreadsheet?

What if some of the biological information could eventually be converted into a simple colour response?

This is where my longer-term research vision comes in.

Imagine a small disposable strip containing several reaction zones.

Each zone could respond to a selected microbial or metabolic indicator.

As the reactions occur, the zones change colour.

An investigator could then photograph the strip using a smartphone.

The workflow could eventually look like this:

Biological sample

↓

Chromogenic sensor

↓

Colour pattern

↓

Smartphone camera

↓

Computer vision

↓

AI interpretation

↓

Decomposition-stage profile

This would transform a complex biological measurement problem into something much more accessible.

From portable biosensing to smartphone-assisted analysis. Examples of approaches that combine portable biosensors, colourimetric or optical responses, smartphone imaging and machine-learning analysis demonstrate the broader technological foundation for developing accessible field-oriented sensing systems.https://link.springer.com/article/10.1186/s13054-023-04365-1

The smartphone would not simply be taking a photograph.

It could become the interface between a physical biological sensor and an AI interpretation system.

The Google AI connection

This is where the project becomes particularly interesting as a builder challenge.

The objective is not simply to attach "AI" to an existing forensic research project.

The objective is to explore how Google AI technologies can help turn scientific observations into a practical prototype.

There are several potential layers.

1. Pattern recognition

The AI can help explore relationships between microbial and environmental variables.

2. Classification

Experimental observations can potentially be grouped into different decomposition-stage profiles.

3. Computer vision

If the future sensor produces colour changes, computer vision can be used to quantify colour intensity and spatial patterns.

4. Multimodal interpretation

Eventually, the system could combine image-derived information with environmental and contextual variables.

5. A field-facing interface

The final goal would be to present complex biological information in a form that is understandable and useful to a forensic investigator.

The technology therefore becomes a bridge between laboratory science and field investigation.

Why start locally?

One of the reasons I find this project meaningful is that the starting point is not an abstract global dataset.

The experimental work is being developed under conditions relevant to the environment in which the research is being conducted.

Decomposition is influenced by environmental factors. Temperature, humidity, soil and microbial ecology can all affect biological processes.

That means locally generated experimental data can be valuable.

A model developed entirely from data generated in another environment may not necessarily behave in the same way under Indian conditions.

So the broader vision is not simply:

Build an AI model.

It is:

Build a research pathway in which locally generated forensic data can be transformed into locally relevant technological solutions.

 What the prototype cannot do—yet

There is an important scientific boundary I want to maintain.

This prototype is not currently a human postmortem interval calculator.

It has not been validated as a forensic instrument for human remains.

It should not be used to provide an operational PMI estimate.

The experimental model is a proof-of-concept platform for investigating whether microbial patterns contain useful temporal information.

Before anything like this could become a forensic tool, it would require substantial additional work:

• larger and replicated datasets;
• controlled experimental validation;
• testing across different environmental conditions;
• independent datasets;
• identification of the most informative biological markers;
• development and validation of chromogenic reactions;
• smartphone image standardisation;
• robust AI model validation; and
• eventually, carefully designed human forensic research.

That may sound like a long road.

It is.

But that is also what makes the project interesting.

From a digital prototype to a physical device

I see the current AI prototype as only the digital half of a much larger research programme.

The pathway I envision is:

Experimental decomposition research

↓

Microbial dataset

↓

AI-assisted decomposition profiling

↓

Identification of discriminating microbial/metabolic markers

↓

Chromogenic sensor development

↓

Smartphone-based colour analysis

↓

AI interpretation

↓

Independent validation

↓

Potential forensic application

The eventual objective is not to make an impressive demonstration.

It is to investigate whether a scientifically grounded, low-cost and field-oriented approach to decomposition assessment is possible.

What building this taught me

Perhaps the biggest lesson has nothing to do with algorithms.

It is that building an AI solution begins with defining the problem correctly.

It would have been easy to start with:

"How can I use AI in forensic science?"

Instead, I started with:

"What information do forensic investigators need, and why is obtaining it difficult?"

That changed everything.

The AI became a means rather than the destination.

The scientific question came first.

The experimental data came next.

Then came the prototype.

And only after that did the larger technological vision begin to emerge.

What comes next?

I don't know yet whether this idea will ultimately become a validated forensic technology.

That is not a weakness of the project.

It is the reason for building the prototype.

Today, the concept is:

Data → AI → decomposition-stage profile

Tomorrow, I hope it can become:

Sample → sensor → colour → smartphone → AI → field decision support

And somewhere between those two points lies a great deal of research.

For me, that is what being a builder means.

It is not about claiming that a problem has already been solved.

It is about taking a real problem, using the tools available today to build the first credible solution, testing where it works and where it fails, and then using what you learn to build the next version.

I started with a question about decomposition.

The next step is to see whether we can teach technology to recognise the biological signals hidden within it.

The science comes first. The AI helps us see the pattern. And the goal is to turn that pattern into something useful.