Neu / Index 001AI product design
and development
01 Agentic systems02 Workflow automation03 Model tuning
A trefoil knot made of tightly wound parallel filaments, rendered as a white dot mesh

UROOJ FATIMA

I design and build AI agents, automate the work between your tools and tune models for the tasks your team does every day.

Each project starts with one real task and ends with a system your team can test, correct and run without me.

Explore services Move the cursor over the mesh
(01) Practice

Practice


Neu is the AI product practice of Urooj Fatima. I design and build the agents, workflows and tuned models behind a product, and the screens people use to check their work.

Most AI projects stall between a promising demo and a system a team can trust. The gap is rarely the model. It is the missing rules: what the system may change, where it should stop, and how a person reviews its work.

I close that gap with small, complete versions built on your real inputs. You see the system succeed, fail and recover before anyone decides to roll it out.

Focus
Agents, automation, tuning
Format
Sprints, builds, support
Output
Working code and tests
(02) Services

Three ways to put a model to work.


Service 01 / 03

Agentic systems

I build AI agents that can look up information, choose which tool to use next, and hand a decision to someone on your team.

An agent is useful when the next step depends on what it finds. It reads a request, gathers the records it needs, picks a tool and proposes an action. The hard part is not the model. It is deciding what the agent may touch, where it must stop and how a person checks its work.

  • +A connected agent
  • +A review screen
  • +Test conversations
Typical length4 to 8 weeks
TagsAgents, Tool use, Human review
Read the service
Service 02 / 03

Workflow automation

I connect the forms, inboxes and apps your team already uses so routine work moves between them automatically. People see the exceptions that need a decision.

Most teams lose hours each week copying data from one tool to another and fixing the records that did not match. A good automation does the routine part quietly and makes the exceptions obvious.

  • +A working integration
  • +An exception queue
  • +A run history
Typical length2 to 6 weeks
TagsIntegrations, Exceptions, Run history
Read the service
Service 03 / 03

Model tuning

I adapt a language model to your categories, terminology or output format, then compare it with the original on examples it has never seen.

A general model knows a lot about the world and very little about your categories. Prompting closes part of that gap. Tuning on examples labelled by your team can close more of it, but only a fair comparison tells you whether it was worth the effort.

  • +A baseline and test set
  • +A tuned model or adapter
  • +An evaluation and handover
Typical length3 to 6 weeks
TagsFine-tuning, Evaluation, Adapters
Read the service
(03) Lab

Small working demos you can try.


Each service page includes one decision from that service, built to run in your browser. Nothing you type leaves the page.

(04) Approach

How a project runs.


Every engagement follows the same four moves. The order matters more than the tools: we look at the real work before anything is built, and we test the awkward cases before anything is released.

About Urooj
  1. 01

    Map the work

    Walk me through a recent task from the first request to the final decision. We trace the tools, the people involved and the points where someone has to exercise judgment.

    You getA task map and a list of decisions
  2. 02

    Build a working version

    I build one complete path through the task using representative inputs. Your team can try it, correct the output and see exactly where a person steps in.

    You getA working prototype on real inputs
  3. 03

    Test the edges

    We run ordinary cases alongside the awkward ones: missing records, failed lookups, unclear requests. Each failure becomes a test that stays in the suite.

    You getA test set and a failure log
  4. 04

    Hand over

    We agree on the conditions for release. Your team gets the setup, the test examples and a guide to making changes without me in the room.

    You getCode, tests and a written guide
(05) Engagements

Pick a starting format.


F/012 weeks

Discovery sprint

For teams that know the problem but not the shape of the solution. We map one task, test a quick prototype and decide what is worth building.

  • Task map
  • Clickable or runnable prototype
  • Build plan with risks
F/024 to 8 weeks

Build

For a defined task that is ready to automate. I design and build the system, the review screen and the tests, then hand it over.

  • Working system
  • Review interface
  • Test suite and guide
F/03Monthly

Ongoing support

For systems already in use. I review failures, extend the test set and adjust prompts, rules or models as your work changes.

  • Monthly review
  • Test set updates
  • Priority fixes
(06) Questions

Before you ask.


Yes. I start with the apps, data and accounts you already use and add new software only when it is the simplest path.

The one that passes the tests for your task. That can be a hosted model from a major provider or an open-weight model you run yourself.

One task, a handful of real examples and an hour with the person who does the work today.

Yes. I can build the first version and pair with your team, or review and extend a system you already have.

You do. Code, prompts, test sets and documentation are handed over at the end of the engagement.

(07) Start

Start with one task.


Write a four-line brief here. It stays in your browser until you copy it. Bring it to a first call and we can map the work in an hour.

01 / Service
04 / Timing
brief.txt67 chars
NEU / PROJECT BRIEF
Service:  -
Task:     -
Tools:    -
Timing:   -
Describe the task to continue