Manish Sharma Lab / Industrial AI & decisions
Menu

I build AI systems for industrial decisions that need evidence, not just predictions.

I build evidence-aware AI systems for industrial decisions. My deepest public work is in Laser Metal Deposition and Directed Energy Deposition at Exafuse, where signals, models, engineering constraints, and human judgment must resolve into a defensible next action.

Run the Decision Cockpit View public proof

Industrial AI / decision systems. Evidence-aware by design. Read the method before using a prediction as a decision.

Melt pool Created in the laser metal deposition process.

Public proof

A public LMD case, a working decision product, and an authored evidence boundary.

Start with physical context, then inspect the decision structure and the limit attached to each source.

Read the proof story
Large LMD-manufactured bridge node component from the Duisburg project
Exafuse — Duisburg bridge components Large structural LMD proof component from the Duisburg bridge story. View the Exafuse source

750+ kgDocumented components

6 nodesStructural nodes

219 hSingle-node build

Monitoring and project scale are public context—not engineering approval or a transferable feasibility claim.

Working product

LMD Decision Cockpit

Make the evidence burden, critical gaps, risk, and next action explicit before an industrial handoff.

Open the preview

Authored boundary

Camera Is Not a Certificate

Working decision product

See what the decision needs before asking for confidence.

The Cockpit keeps the decision signal, critical gaps, risk, evidence needed, and next action visible. Confidence is not approval.

Active module

Decision Cockpit

  • Local session
  • Browser-local
  • Decision-support only
LMD Decision CockpitLMD Decision Brief v1.0Conservative output

Start with a rough LMD question. Leave with a brief.

Pick the situation, mark what is known, then expose missing information, risk flags, evidence needed, and an Exafuse review route. Inputs stay in this browser session only.

  • ;
  • ;

Example scenario: worn steel shaft near bearing seat.

Review the worked output, start your own brief here, or open the full workbench.

Repair damaged/worn part: start with LMD Repairability Quick Check. Not enough information. Missing information and risk flags remain visible in the brief.
Start your own brief Open full brief

Sense → Model → Decide → Verify

To me, a prediction is only one part of the job. The next action, the missing context, and the verification path have to be clear too.

  1. 01 Sense Capture the process signal and its operating context.
  2. 02 Model Make assumptions visible.
  3. 03 Decide Plan the evidence needed.
  4. 04 Verify Use inspection evidence.

The broader question behind the current work

I am studying how this approach can help with decisions based on incomplete physical signals, operational risk, and human responsibility. That is a research direction, not a cross-industry deployment claim.

  • Incomplete physical signals
  • High cost of false confidence
  • Model output connected to inspection
  • Operational context changes the recommendation
  • Human responsibility remains explicit
  • Traceability matters after the model runs

Selected work

One public industrial story, supported by a working product and an authored note

The opening proof is physical and source-backed. This supporting industrial case, the Cockpit, and the note show how I structure a decision and its evidence boundary.

LMD Decision Cockpit

Problem
A decision starts before every fact is known.
Helps decide
Turn facts, gaps, risk, and evidence needs into a next action.
Cannot prove
Replace engineering approval or final feasibility review.

Camera Is Not a Certificate

Problem
A process image can look convincing before it becomes sufficient evidence.
Helps decide
Separate monitoring signals from an inspection-backed quality claim.
Cannot prove
Turn a signal into a release decision.
Explore all frameworks

Authored note

One idea behind the broader platform

A first-person note on why model output, context, action, and verification need to stay connected.

All lab notes RSS feed

A Prediction Is Not Yet an Industrial Decision

Why a model output needs context, a clear next action, and a verification path before it becomes useful industrial decision support.

Personal platform

Explore the method, current work, and research questions.

Public frameworks, tools, and notes on Industrial AI & Decision Systems, grounded in current LMD/DED work at Exafuse.

Explore my work About Manish

Commercial boundary

For commercial LMD/DED services and RFQs, use Exafuse.

My current applied LMD/DED work is carried out through Exafuse. This site shares public methods and notes; company services, case studies, and engineering review belong there.

Request Exafuse review Contact routes

Personal mission

Build products that make difficult industrial decisions clearer without hiding uncertainty.

I am building toward products and ventures that help people make difficult industrial decisions without hiding uncertainty or replacing engineering responsibility. LMD/DED is where I am proving the method in public today.

Discuss an industrial AI decision problem

Suitable for evidence-aware AI workflow design, monitoring and verification architecture, industrial decision-support products, human-in-the-loop engineering systems, and research or product collaboration.