Your environment
Container-first and cloud-native, available for deployment in your cloud, on-premise, or in a sovereign environment where residency requires it.
lalla.ai — the Innopas AI platformBuilt once. Every product runs on it.
lalla.ai is the AI platform beneath everything Innopas ships: graph-grounded retrieval, agents, evidence, evaluation, guardrails, and Lalla Chat as the surface people actually touch. A new product brings a domain graph and an interface. It does not bring its own AI stack.
Surface
Reference trace
Schematic of the request path. Illustrative, not live traffic.
The Innopas model
Research proves it once. The platform carries it. Products pay only for their domain.
Innopas is an AI technology company. We build our own AI platform, our own products, and the research that makes them hard to copy. lalla.ai is the middle link, and it is the one that compounds.
Layer 01 · Research
DTRIHub
DeepTech Research & Intelligence Hub. Representation, forecasting, attribution and explainability, proven against a baseline before anything ships upward.
→
Layer 02 · Platform
lalla.ai
Models, retrieval, agents, evaluation, guardrails and Lalla Chat. Built once. Shared by everything above it, and versioned so a product team cannot quietly fork it.
→
Layers 03–04 · Domain & product
The products
A domain graph and the screens people live in. EUNIQ for energy and utility networks, TopSyllabus for education. Nothing below layer three is rebuilt.
The test of a platform is the second product. Anything can be called a platform while one product runs on it. TopSyllabus and EUNIQ share no customers, no regulator and no vocabulary. They run on the same layer two.
The stack, live
Everything above layer two is a thin layer.
Five planes, wired as they actually run. Research feeds the platform from underneath; the six capabilities cross-wire to each other; two domain graphs sit on top of them; and the products are the thinnest thing in the diagram.
Around all of it, four solutions hold their own orbits: DeepTech, Data Spine, AI Engine and Cloud Mesh. They are the ways a client enters, and each one draws on the same core. Drag to turn it.
lalla.ai · platform topology · drag to turn
platform & research
utility branch
education branch
not built yet
DeepTech
Data Spine
AI Engine
Cloud Mesh
Five layers, bottom to top: DTRIHub research; the lalla.ai platform core with six capabilities; the utility knowledge graph and concept graph; the EUNIQ and TopSyllabus products; and the surfaces: Lalla Chat, dashboards and APIs, field and operations. Four solutions orbit the core: DeepTech, Data Spine, AI Engine and Cloud Mesh.
Topology, not telemetry. The wiring is how the platform is composed; the pulses are illustrative.
Inside the platform
What layer two actually provides.
Six capabilities, versioned and shared. A product team consumes them; it does not reimplement them. Each one exists because something specific goes wrong without it.
P-01 · Grounding
Graph-grounded retrieval
Retrieval that walks a typed domain graph rather than searching a pile of documents. The answer is assembled from the structure, so it can name the node it came from.
Why Flat retrieval cannot tell you which transformer, or which prerequisite.
Typed graph walkHybrid retrievalEntity resolutionDocuments & telemetry
P-02 · Reasoning
Agents and tool use
Agents that query the graph, run a model, call a service and compose a result. Each step is recorded, so what happened can be reconstructed later rather than inferred.
Why An unexplained action is one nobody in a regulated business will authorise.
23 agentsTool callingStep recordingHuman-in-the-loop
P-03 · Evidence
Evidence and confidence
Every output carries its supporting evidence and a confidence value, produced at the same time as the answer rather than reconstructed afterwards.
Why This is what a regulator, an auditor or a parent is actually asking for.
ProvenanceConfidence valuesNamed nodesAudit log
P-04 · Evaluation
Evaluation harness
Test sets, baselines and regression runs for every capability. A change ships when it beats the previous version on the record, not when it demos well.
Why Without this, a model upgrade is a gamble taken in production.
Test setsBaselinesRegression runsDrift tracking
P-05 · Guardrails
Guardrails and policy
Access control down to the graph node, redaction, refusal behaviour and audit logging. A student and an operator see only their own scope, enforced below the product rather than inside it.
Why Permissions implemented per product are permissions implemented inconsistently.
Node-level accessRedactionRefusal policyRetention
P-06 · Portability
Model and deployment choice
Model-neutral by design, and deployable in your cloud, on-premise or in a sovereign environment. Some customers cannot send data anywhere else.
Why A platform locked to one vendor or one region is a platform with a shelf life.
Model routingMultiple model familiesPrivate & on-premiseSovereign deployment
Lalla Chat
The same component, grounded two different ways.
Lalla Chat is the same component in both products. What changes is the graph it is grounded in and the scope it is allowed to see. Ground the answer in a graph, name the node, show the evidence, state the confidence. Only the vocabulary belongs to the industry.
Use the surface switch above to move the page between the two. Nothing about the platform changes.
Class 9
Why did I get transpiration wrong again?
Lalla
Walking back from that question through its prerequisites: across four answers you describe the effect but never the driver.
The gap is one concept upstream — water potential. Fix that and the transpiration questions stop being memory work.
The gap is one concept upstream — water potential. Fix that and the transpiration questions stop being memory work.
Evidence: 4 answers · concept graph
Class 9
Show me the four.
Lalla
Opened, with the sentence in each one where the driver should have appeared. Then three questions on water potential before you retry.
Control room
Why did losses jump on feeder F-09?
Lalla
Walking the graph from feeder to transformer to meter: at DT-024, metered consumption fell while load held steady.
That pattern is the one worth a crew. Sibling transformers on F-09 show nothing like it.
That pattern is the one worth a crew. Sibling transformers on F-09 show nothing like it.
Evidence: DT-024 · 30-day window · confidence 0.81
Control room
Open the node.
Lalla
DT-024 open, with the six meters behind the divergence and the evidence window attached to the case.
Ask Lalla anything in this syllabus
Running on layer two
Two domains that share nothing, except this.
One core, many domains. Every product sits on the same platform and adds only the knowledge of its own field, which is why the second product cost a fraction of the first, and why a competitor has to rebuild the foundation rather than copy the interface.
Energy & utility
EUNIQ
Electricity, gas and water network intelligence. Loss, assets, forecasting and field work, with Cockpit, Command and Field for utility teams.
euniq.ai ↗ EducationTopSyllabus
AI exam practice with evidence behind every score. Concept mastery, evaluated answers, and dashboards for students, parents and schools.
topsyllabus.ai ↗ The companyInnopas
The AI technology company behind both: the products, the platform underneath them, and the research programme that keeps the platform ahead.
innopas.com ↗Runtime & governance
Built for the rooms where the data cannot leave.
Both of our domains are governed ones. The platform is designed so that a security review is a conversation about architecture rather than a negotiation about exceptions.
Where it runs
Scope
Enforced below the product
Access control to the graph node, tenant isolation and retention rules implemented in the platform, so every product inherits the same behaviour.
Traceability
Every answer explains itself
What was retrieved, which agent ran, which model answered and what confidence it carried, all recorded with the answer rather than reconstructed after it.
Interoperability
Open at every layer
APIs into retrieval, agents and evaluation, so the platform sits inside your stack rather than replacing it. License it, or have us build your domain on it.
DTRIHub
The research that makes it hard to copy.
DTRIHub, our DeepTech Research & Intelligence Hub, is a standing team rather than a project allocation. It works with university research groups and our PhD advisors on representation, forecasting, attribution and explainability. A method is proved against a baseline in research before it is allowed into the platform.
23+Agents in the library
4Proprietary algorithms
3Active research domains
- Energy & utilityTopology-aware forecasting, multi-signal loss attribution and evidence-first case generation over network graphs.
- EducationModelling what a learner has actually misunderstood, rather than which questions they got wrong.
- PaymentsIn research with DTRIHub. Anomaly detection at volume, anchored by domain experience at the level where national payment infrastructure gets designed.
- Your domainBring the domain and the data. Everything from layer two down already exists.
Questions
The six we get asked first.
What is lalla.ai?
lalla.ai is the AI platform Innopas builds on: layer two of a four-layer stack that runs research, platform, domain and product. It provides graph-grounded retrieval, agents, evidence and confidence, an evaluation harness, guardrails and model portability. A new product brings a domain graph and an interface; it does not bring its own AI stack.
Is lalla.ai a wrapper around a large language model?
No. It is model-neutral by design and routes across model families. Retrieval walks a typed domain graph rather than searching a pile of documents, scope is applied before anything is read, and every answer carries the evidence and confidence it was produced with.
What is Lalla Chat?
Lalla Chat is the conversational surface of the platform, and it is the same component in every product. What changes between products is the graph it is grounded in and the scope it is allowed to see, not the component itself.
Which products run on lalla.ai?
EUNIQ, an intelligence platform for electricity, gas and water networks, and TopSyllabus, an exam practice platform for K-12 education. The two share no customers, no regulator and no vocabulary, and both run on the same layer two.
Where can lalla.ai be deployed?
It is container-first and cloud-native, and is available for deployment in your cloud, on-premise, or in a sovereign environment where data residency requires it. Access control is enforced to the graph node, below the product rather than inside it.
Can lalla.ai be licensed for another domain?
Yes. The platform exposes open APIs into retrieval, agents and evaluation, so it can sit inside an existing stack. Innopas can also build a domain on it. The starting point is a thirty-minute call with the engineers who would do the work.
Start here
Bring one problem. We'll tell you if it's worth building.
Thirty minutes with the engineers who would do the work. No deck, no discovery invoice. A straight read on feasibility, sequence and what a first build would take, including when the answer is that you should not build it.
30 min · video call · Who joins · engineering, not sales · Cost · none
Emailreachus@innopas.com
Phone+91 44 4904-2818
CompanyInnopas ↗
PresenceNorth America · Asia Pacific