01
Outpatient claims grouper
A classification model over outpatient billing data: identifies which lines form a single visit, assigns it to one of 128 payment classes, and records the rule applied. Ambiguous rows are referred to a coder.
Turn the problem until it looks simple.
- Products
- Internal systems
- Data
- AI
The studio
Most complexity is just the wrong angle.
Oblimin is named after an oblique rotation: a method for turning a mess of correlated measurements until a simple structure appears. The data stays as it is. Only the angle changes.
i
Evidence over assumption
Our approach is evidence-based: we observe how a product performs in real use, measure where it succeeds and where it fails, and let that direct what comes next.
ii
We advise, we build, we stay for the long run
We work across the whole arc of making software: product, design, engineering, the data underneath, and the work that keeps it running once it is live.
How we work
Four moves. One continuous team.
the field as it arrives, nothing arranged yet
Scope
Establish the problem.
A short study of the people involved, the systems already in place and the constraints that cannot move. It ends in a written problem statement, not a proposed solution.
Prototype
Test the uncertain part first.
The part of the system carrying the most uncertainty is prototyped ahead of the rest, on real data where it exists. What it settles determines the scope of the build that follows.
Build
Deliver in usable stages.
Each stage produces something usable on its own. Work carrying no external dependency is delivered first, so software is in use while access and agreements are still being arranged.
Operate
Run it after release.
Hosting, monitoring and the changes that follow real use. Where a system is intended to act on its own, that is enabled case by case on measured evidence.
Capabilities
What we build, and how far we go.
Six areas of work. Most engagements draw on more than one, and the useful question is usually not which area a piece of work sits in but how far along it we are asked to go.
Hover a point to read more.
- Websites & web applicationsMarketing sites, portals, dashboards and full web applications.
- Custom software & internal toolsInternal dashboards, admin panels, and integrations between systems that share data.
- AI systems & integrationLLM features, retrieval systems and agents, inside an existing product or workflow.
- Product & MVP developmentAn idea to a first working version: scoping, design, build and release.
- Workflow automationData entry, report generation, file handling, and moving information between tools.
- ResearchFeasibility studies, prototypes and technical assessments.
Three modes
The matrix is deliberately porous. A project can move from research to build to operation without changing hands.
Advise
Make the decision clearer.
Evidence, feasibility, scope and technical direction before commitment.
Build
Make the idea real.
Design, engineering and the first working version, built by the same team.
Operate
Keep it useful.
Monitoring, maintenance and changes driven by what real usage reveals.
Our work
Built, and still running.
Analytics pipelines, LLM systems and agents delivered across healthcare, retail, media and education. Client names and identifying detail are withheld by agreement.
02
Patient history assistant
A retrieval-grounded LLM assistant over the patient record: answers clinical queries and cites the specific note each answer is drawn from.
03
Agentic commerce operations
Market analytics end to end: collection pipelines, models applied to the data, and decision outputs. Agents monitor product lines across marketplaces and act on stock, price and promotion, with no change made without approval.
04
Voice for a storytelling platform
Voice cloning and text-to-speech models in an Indian language, built as a pipeline that accepts new voices without being rebuilt for each one.
05
Study assistant over textbooks
A retrieval-grounded LLM tutor over textbook content: every answer displays the source passage rather than relying on the model’s general knowledge.
06
Virtual try-on
Image models that render how a garment appears on a shopper from standard phone photographs, with the full pipeline running on self-hosted infrastructure.
Engagement
Start small. Grow quickly.
What we ask for
Access to the people who do the work, a sample of the real data, and a named person on the client side who can settle a question. Long-lead items are identified in the first week rather than the last.
What we hand over
The source, the deployment, the documentation, and whatever runs it. Nothing is held back to secure a retainer: continuing is the usual outcome rather than the contracted one.
Contact
Change the angle.
A short description of the problem is enough to start. Every message is read and answered by the people who would do the work, usually within a working day.
Prefer plain email? connect@oblim.in