# CalQuity Labs
> CalQuity Labs builds AI around your business to reduce repetitive work, improve decisions and help you serve customers better, with an embedded engineering team.

AI-native products and forward-deployed engineering for business teams. Contact: hello@calquity.com.

## Markdown versions
Every public content page has a Markdown twin generated from its rendered content. Append .md to the page path; the homepage is /index.md. HTML remains the canonical version.
- [Markdown directory](https://calquitylabs.com/markdown-index.md)
- [Complete text](https://calquitylabs.com/llms-full.txt)
- [Homepage Markdown](https://calquitylabs.com/index.md)

## Solutions
Business workflows and local model deployment apply across industries; requirements are assessed for each workload.
- [SLMs & local models](https://calquitylabs.com/solutions/slms-and-local-models): Run AI on company servers or devices. We select, adapt and deploy models for your tasks, with privacy controls and speed and operating cost tested at your expected volume.
- [Workflow automation](https://calquitylabs.com/solutions/document-systems): Fill forms, create and send invoices, update records and complete routine tasks across the tools your team already uses. Agree what can run automatically and route exceptions to the right person.
- [Support & voice](https://calquitylabs.com/solutions/support-and-voice): Answer an order query, book an appointment or log a service request through phone or chat. Connect the conversation to your records, and pass complex cases to your team with context.
- [Planning & coordination](https://calquitylabs.com/solutions/operations-and-planning): Bring schedules, stock, orders and deliveries into a shared plan. See what is blocked, compare options and coordinate the next steps with the people responsible.
- [Company knowledge](https://calquitylabs.com/solutions/private-ai): Find a policy, recover a project decision or ask about an internal record. Search connected documents, messages and tools in one place, with source references and your existing access permissions.

## Industries
Each industry page combines relevant solutions and includes contextual use cases, proposed changes and review owners. Applications describe possible work, not delivered client projects.
- [Manufacturing & distribution](https://calquitylabs.com/industries/manufacturing): Fewer order corrections, earlier warning of planning problems and faster answers for customers.
- [Logistics & freight](https://calquitylabs.com/industries/logistics): Less time chasing shipments, document gaps caught sooner and customers kept informed.
- [Financial services](https://calquitylabs.com/industries/financial-services): More analyst time for judgement, and routine operations and service work resolved sooner.
- [Property & leasing](https://calquitylabs.com/industries/property): Renewal deadlines kept in view, less lease admin and tenant requests that keep moving.

## Working with CalQuity Labs
### What does a forward-deployed AI engineering team do?
A forward-deployed AI engineering team works alongside the people who run a business workflow. CalQuity Labs combines product and software engineering with AI integration, taking a problem from discovery through a working system, deployment and ongoing improvement.

### What kinds of AI systems does CalQuity Labs build?
We build and deploy SLMs and local models, workflow automation, support and voice systems, planning and coordination tools, and company knowledge assistants. The starting point is the business task, its users and the data available—not a fixed catalogue of agents.

### Can you work with our existing software and data?
Yes. We design around existing tools, data sources and permissions. Early work establishes which integrations are available, what data needs preparation and where people should review outputs or handle exceptions.

### Can an AI system run on our own infrastructure?
Yes. On-premise, private-cloud and edge environments can be part of the design when data or operating constraints require them. Model choice, hardware, connectivity, access controls and maintenance needs are assessed before the deployment approach is agreed.

### How do we decide whether an AI workflow is worth building?
Start with a repeated task, a measurable bottleneck and representative examples. Together, we examine the inputs, acceptable error rates, review requirements and integration effort. Success measures can include time saved, turnaround time, exception rates or the quality of decisions supported.

### What should we bring to the first conversation?
Bring a workflow, product idea or operational problem, along with the people who know it well. Examples of current inputs and outputs, the tools involved and a description of what a better result would look like help make the conversation concrete. Scope and delivery plans follow that assessment.

The solution pages describe possible workflows and their implementation requirements.

## Insights

### Case Studies
Client projects and product case studies: https://calquitylabs.com/insights/case-studies.
- [One portal for every lease, bill and repair.](https://calquitylabs.com/insights/case-studies/warehouse-tenant-portal): How CalQuity brought leases, rent billing, deposits and property upkeep into one shared portal for a warehouse portfolio operator.
- [From manuscript to publication-ready files.](https://calquitylabs.com/insights/case-studies/editorial-workflow-automation): How CalQuity Folio brings manuscript cleanup, reference checks, tracked changes and validated XML export into one editorial workflow inside Word.

### Blog (44 posts)
Engineering posts on building AI systems that survive production. Index: https://calquitylabs.com/insights/blog.
Articles discuss the limits of their approach and link citations beside the claims they support.
How to read these posts, covering what the figures are and how sources are checked: https://calquitylabs.com/insights/disclaimer.

### Private, on-premise and sovereign AI
[Hub](https://calquitylabs.com/insights/private-ai): Private AI is defined by control you can prove: who can access the system, where data moves, what gets retained, and how copies are deleted. Running a model on your own hardware answers only the hosting question.

- [What makes an LLM private?](https://calquitylabs.com/insights/private-ai/makes-llm-private): Privacy is established by evidence about access, retention, training use, egress, logs, keys, and deletion, not by a hosting label. (published 2026-07-10)
- [How to preserve RAG access controls end to end](https://calquitylabs.com/insights/private-ai/preserve-rag-access-controls-end-end): Enforce access rules across indexes, embeddings, caches, logs and evaluation data. Test permission changes at every retained copy, not only the user-facing query. (published 2026-08-11)
- [Local LLMs are not automatically HIPAA compliant](https://calquitylabs.com/insights/private-ai/local-llms-automatically-hipaa-compliant): Running a model on your own hardware satisfies none of the Security Rule by itself. The obligations attach to access control, audit logging, integrity, transmission security and the agreements covering anyone who touches the data, and self-hosting changes where those controls live rather than whether you need them. (published 2026-08-27)
- [How much hardware does your on-premise LLM need?](https://calquitylabs.com/insights/private-ai/much-hardware-premise-llm-need): Size hardware from in-flight model requests, context length and response deadlines. Test both memory capacity and serving latency before pricing the configuration. (published 2026-08-31)
- [Does our data have to leave the building?](https://calquitylabs.com/insights/private-ai/does-our-data-have-to-leave-the-building): No, but keeping it in-house costs hardware, power and a person, and in-house is a claim to be proved, not a label to be accepted. (published 2026-09-24)
- [How to prepare company documents for AI search](https://calquitylabs.com/insights/private-ai/prepare-company-documents-for-ai-search): Choose authoritative sources, assign owners and test permissions and updates before expanding internal AI search. (published 2026-09-29)

### AI agents that survive production
[Hub](https://calquitylabs.com/insights/agents): An agent becomes an operating risk when it can change external state. A retry can duplicate a payment, a document can distort tool instructions, and a vague approval can authorize the wrong action.

- [Chatbot, copilot, workflow, or AI agent?](https://calquitylabs.com/insights/agents/chatbot-copilot-workflow-ai-agent): Classify systems by the external state they may change and increase permission, approval, and recovery controls with consequence. (published 2026-07-20)
- [How to govern AI agent traces used for evaluation](https://calquitylabs.com/insights/agents/govern-ai-agent-traces-used-evaluation): Traces kept for evaluation are a second production dataset carrying the same access, retention and consent obligations as the first. Teams govern the primary store and leave the trace store open, so the copy made for measurement becomes the weakest point. (published 2026-08-13)
- [Your AI agent needs a safe way to stop](https://calquitylabs.com/insights/agents/ai-agent-needs-safe-way-stop): Each failure class must choose refusal, deterministic fallback, bounded partial output, or a human queue instead of forcing every uncertain exit into one shape. (published 2026-08-20)
- [What must an AI support handoff preserve?](https://calquitylabs.com/insights/agents/ai-support-handoff-preserve): A handoff is complete only when verified identity, conversation, evidence, attempted actions, tool results, unresolved work, and risk flags reach an owned queue. (published 2026-08-25)
- [Which customer support requests should you automate first?](https://calquitylabs.com/insights/agents/which-customer-support-requests-to-automate): Start with requests that have an approved answer, accessible records, bounded authority and a staffed fallback. (published 2026-09-29)

### Custom models, fine-tuning and evaluation
[Hub](https://calquitylabs.com/insights/custom-models): Custom training starts with diagnosis. Missing evidence calls for retrieval, inconsistent behaviour may call for prompting or training, and a fixed business rule belongs in ordinary software.

- [How much training data does your custom AI model need?](https://calquitylabs.com/insights/custom-models/much-training-data-custom-ai-model-need): The stopping point comes from stratified learning curves, uncertainty, marginal labeling cost, and alternatives, not a reusable example count. (published 2026-08-24)
- [Require vendors to prove every custom model can be rebuilt](https://calquitylabs.com/insights/custom-models/require-vendors-prove-every-custom-model-rebuilt): A signed manifest proves integrity and traceability of recorded inputs; rebuildability additionally requires an executed reproduction test and declared tolerance. (published 2026-09-01)

### Documents and workflow automation
[Hub](https://calquitylabs.com/insights/documents): Document work can support a decision or complete a task, from filling a form to preparing an invoice. These guides cover source records, access boundaries, exceptions and the review needed before the next action.

- [How enterprise RAG works across two pipelines](https://calquitylabs.com/insights/platform/enterprise-rag-works-across-two-pipelines): Enterprise RAG is two pipelines running on different clocks, an ingestion pipeline that builds and refreshes the index and a query pipeline that serves. Most failures live in the gap between them, and treating retrieval as one request path hides all of them. (published 2026-07-23)
- [How to preserve RAG access controls end to end](https://calquitylabs.com/insights/private-ai/preserve-rag-access-controls-end-end): Enforce access rules across indexes, embeddings, caches, logs and evaluation data. Test permission changes at every retained copy, not only the user-facing query. (published 2026-08-11)
- [How much of my team's time will checking the AI take?](https://calquitylabs.com/insights/deciding-what-to-build/how-much-time-will-checking-the-ai-take): Compare preparation, review, correction and sign-off time with the manual baseline. Mandatory review can still save time; unattended processing needs its own evidence and approval policy. (published 2026-09-08)
- [How to evaluate AI invoice processing before ERP integration](https://calquitylabs.com/insights/documents/evaluate-invoice-processing-ai): Judge invoice automation by correctly accepted records, a usable exception queue and reliable ERP posting. (published 2026-09-29)

### Equipment, maintenance and control
[Hub](https://calquitylabs.com/insights/vision): Maintenance and equipment decisions need evidence from the actual installation. These articles examine operating cost, failure modes and the boundaries between AI advice and equipment control.

- [Does predictive maintenance AI actually save money?](https://calquitylabs.com/insights/vision/predictive-maintenance-ai-actually-save-money): For one advisory asset, price the deterministic safety boundary, errors, review, fallback, integration, and maintenance without implying that AI should control equipment. (published 2026-08-17)
- [Keep generative AI outside deterministic control loops](https://calquitylabs.com/insights/vision/keep-generative-ai-outside-deterministic-control-loops): A generative model may advise an operator or draft a change, but it must sit outside the loop that actuates equipment. The protection layer has to stay deterministic, independently verifiable and separately certified, because a probabilistic component cannot carry a safety integrity claim. (published 2026-09-03)

### Serving, observability and AI operations
[Hub](https://calquitylabs.com/insights/platform): Retrieval selects the evidence, serving determines waiting time, caching shapes the bill, and telemetry decides whether a failure can be understood. These layers often matter more than the model name.

- [How enterprise RAG works across two pipelines](https://calquitylabs.com/insights/platform/enterprise-rag-works-across-two-pipelines): Enterprise RAG is two pipelines running on different clocks, an ingestion pipeline that builds and refreshes the index and a query pipeline that serves. Most failures live in the gap between them, and treating retrieval as one request path hides all of them. (published 2026-07-23)
- [Write a disaster recovery plan for your AI system](https://calquitylabs.com/insights/platform/write-disaster-recovery-plan-ai-system): Recovery objectives must be stated separately for the index, the weights, the trace store, and the queue, because each has a different rebuild time and a different data loss cost. (published 2026-08-18)

### Forward-deployed engineering, explained
[Hub](https://calquitylabs.com/insights/forward-deployed): A forward-deployed engineer works beside the people whose workflow is changing. They observe the job, connect the surrounding systems, and stay accountable through deployment.

- [What does a forward-deployed engineer actually do?](https://calquitylabs.com/insights/forward-deployed/what-a-forward-deployed-engineer-does): A forward-deployed engineer works directly with business users and owns the changes needed to make a system work in their operation. Access to the workflow and responsibility for delivery shape the role; on-site and remote arrangements depend on the work. (published 2026-09-07)
- [What is forward-deployed engineering?](https://calquitylabs.com/insights/choosing-a-partner/what-is-forward-deployed-engineering): Forward-deployed engineering shortens the path from observing a wrong assumption to changing the system. Direct access to the work and authority to make bounded changes are essential. (published 2026-09-13)
- [How does an engagement with an embedded engineering team start?](https://calquitylabs.com/insights/forward-deployed/how-an-embedded-engineering-engagement-starts): It starts with a conversation about the work rather than a requirements document: we look at what you do, agree what is worth building, scope it, and put engineers on it, on site where that helps. (published 2026-09-16)
- [What will the first weeks with engineers on site ask of my team?](https://calquitylabs.com/insights/forward-deployed/first-weeks-with-engineers-on-site): The first ask is that you walk us through the work as it actually runs, because the fastest way to build something useful is to watch the job before writing anything that claims to do it. (published 2026-09-17)
- [What happens to the system after the first version works?](https://calquitylabs.com/insights/forward-deployed/what-happens-after-the-first-version-works): A working system is never finished, because the work it supports keeps changing; the question that matters is who holds the ability to keep changing it, and that can be your team, ours, or both. (published 2026-09-18)
- [Embedded engineers, an agency, contract staff, or a software licence?](https://calquitylabs.com/insights/forward-deployed/embedded-engineers-agency-contract-staff-or-licence): The four differ on who carries the risk that it never goes live and who knows the work at the end; choose by which of those risks you can afford to hold. (published 2026-09-21)

### Deciding what to build, and whether to build it
[Hub](https://calquitylabs.com/insights/deciding-what-to-build): The first question is whether AI belongs in the workflow. Rules may be enough, usable evidence may be missing, errors may be impossible to contain, or review may cost more than the original task.

- [What does enterprise AI cost after review and rework?](https://calquitylabs.com/insights/deciding-what-to-build/enterprise-ai-cost-review-rework): The useful cost is accepted work after integration, evaluation, review, exceptions, operations, and downstream error, not model usage alone. (published 2026-07-16)
- [Generative AI, predictive AI, and automation solve different problems](https://calquitylabs.com/insights/deciding-what-to-build/generative-ai-predictive-ai-automation-solve-different): The three answer different questions. Automation executes a decision already specified, prediction estimates a quantity from history, and generation produces a candidate artifact that still needs checking. Choosing the wrong one shows up as a verification cost you cannot remove later. (published 2026-07-22)
- [When should a business not use AI?](https://calquitylabs.com/insights/deciding-what-to-build/business-use-ai): Reject AI when rules suffice, representative evidence is unavailable, errors cannot be contained, verification erases value, rights are unresolved, or no fallback exists. (published 2026-07-27)
- [AI governance should produce evidence, not only policy](https://calquitylabs.com/insights/deciding-what-to-build/ai-governance-produce-evidence-only-policy): A governance programme is working only when it produces artifacts an auditor can inspect for a named system on a named date. A policy document proves that someone wrote a policy, which is a different claim. (published 2026-08-06)
- [Measure adoption, not deployment](https://calquitylabs.com/insights/deciding-what-to-build/measure-adoption-deployment): A shipped system that people route around has no value, so rollout needs task-level usage, abandonment, override, and rework measurement alongside model quality. (published 2026-08-10)
- [How much of my team's time will checking the AI take?](https://calquitylabs.com/insights/deciding-what-to-build/how-much-time-will-checking-the-ai-take): Compare preparation, review, correction and sign-off time with the manual baseline. Mandatory review can still save time; unattended processing needs its own evidence and approval policy. (published 2026-09-08)
- [What happens when the AI gets one wrong?](https://calquitylabs.com/insights/deciding-what-to-build/what-happens-when-the-ai-gets-one-wrong): The question for a vendor is not the accuracy figure but what happens to the one it gets wrong: who sees it, when, and what has already happened by then. (published 2026-09-09)
- [Will my staff actually use it?](https://calquitylabs.com/insights/deciding-what-to-build/will-my-staff-actually-use-it): Measure adoption through completed tasks, workarounds and review effort. Fit the system to users’ routines and decisions; logins alone do not establish useful adoption. (published 2026-09-10)
- [Do we need our own AI model?](https://calquitylabs.com/insights/deciding-what-to-build/do-we-need-our-own-ai-model): Test rules and retrieval before custom training. Train when a measurable requirement remains unmet, and compare the complete data, integration, evaluation and maintenance cost. (published 2026-09-14)
- [Do you need AI for this, or just software that works?](https://calquitylabs.com/insights/deciding-what-to-build/do-you-need-ai-or-just-software-that-works): Much of what is sold as AI is a database and a rule; AI earns its place only where the input cannot be specified in advance. (published 2026-09-15)
- [Which job should be the first one you hand to AI?](https://calquitylabs.com/insights/deciding-what-to-build/which-job-to-hand-to-ai-first): The first job is repeated, checkable and boring, with a measure you already report; it is not the most impressive one. (published 2026-09-22)
- [Why do AI demos work and AI systems fail?](https://calquitylabs.com/insights/deciding-what-to-build/why-ai-demos-work-and-systems-fail): A demo runs on chosen inputs with a person steering; a system runs on whatever arrives with nobody watching, and the gap is exceptions, access and an owner. (published 2026-09-23)
- [What will an AI system cost to run after it is built?](https://calquitylabs.com/insights/deciding-what-to-build/what-an-ai-system-costs-to-run): Budget review, corrections, hosting, model usage, integration maintenance and support. Their relative size depends on the workflow; a build-only quote is not a running-cost estimate. (published 2026-09-25)

### Choosing and managing an AI engineering partner
[Hub](https://calquitylabs.com/insights/choosing-a-partner): A convincing demo does not prove that a partner can deliver an operable system. The evaluation needs representative cases, declared failure paths, measurable acceptance, and a handover another team can execute.

- [Can another team operate what your AI vendor built?](https://calquitylabs.com/insights/choosing-a-partner/another-team-operate-ai-vendor-built): Handover passes only when a new team can rebuild, evaluate, deploy, observe, roll back, and operate the system. (published 2026-07-13)
- [What proof should an AI engineering partner provide?](https://calquitylabs.com/insights/choosing-a-partner/proof-ai-engineering-partner-provide): The proof worth asking for is evidence the partner produced as a by-product of doing the work, not documents assembled for the sale: run records, the evaluation set and its failures, the decisions that were reversed, and a handover another team has actually used. Anything authored for the pitch tests writing, not work. (published 2026-07-15)
- [Make AI vendors prove they can run production systems](https://calquitylabs.com/insights/choosing-a-partner/make-ai-vendors-prove-they-run-production-systems): Score representative evaluation, failure handling, deployment evidence, operating ownership, and exit instead of demos, logos, or generic team claims. (published 2026-07-29)
- [An AI pilot must be allowed to fail](https://calquitylabs.com/insights/choosing-a-partner/ai-pilot-allowed-fail): A pilot needs a frozen decision, baseline, representative hard cases, operating exposure, cost boundary, and precommitted no-go criteria. (published 2026-07-30)
- [How should you test an AI vendor before hiring?](https://calquitylabs.com/insights/choosing-a-partner/test-ai-vendor-hiring): Match a paid selection exercise to the delivery model. Test fixed-brief delivery through its returned artifacts, and embedded delivery through the actual workflow, access and approval constraints. (published 2026-08-03)
- [Should you build, buy, or partner for enterprise AI?](https://calquitylabs.com/insights/choosing-a-partner/build-buy-partner-enterprise-ai): Source model, data, workflow, integration, evaluation, experience, and operations independently, with explicit ownership and exit cost. (published 2026-08-05)
- [What should I ask an AI vendor in the first meeting?](https://calquitylabs.com/insights/choosing-a-partner/what-to-ask-an-ai-vendor-in-the-first-meeting): Seven questions, each with what a good answer sounds like and what a bad one hides. (published 2026-09-11)
- [What is forward-deployed engineering?](https://calquitylabs.com/insights/choosing-a-partner/what-is-forward-deployed-engineering): Forward-deployed engineering shortens the path from observing a wrong assumption to changing the system. Direct access to the work and authority to make bounded changes are essential. (published 2026-09-13)

## Company
- About: https://calquitylabs.com/about
- Name: CalQuity Labs
- Legal name: CalQuity Technologies Private Limited
- Registered office: 6 Garstin Place, Suite 501, Kolkata, West Bengal 700001, India
- Email: hello@calquity.com
- CIN: U62099WB2024PTC272030
- Profiles: https://www.linkedin.com/company/calquity/posts/?feedView=all, https://x.com/cqnow_, https://www.youtube.com/@calquity, https://www.calquity.com/

Source: https://calquitylabs.com/