Aptibit

We started an AI company in Kolkata and pointed it at the world. A few years ago that would have sounded ambitious. In 2026 it is just obvious, and here is what changed. The old assumption was that world class technology companies come from a handful of zip codes. Talent everywhere else was for hire, not for building. That assumption is dead, and its death is the whole opportunity. Three things flattened the map. The frontier went public. The research, the models, the tools that define AI reach a team in Kolkata the same day they reach anyone else. There is no proximity advantage to being near a specific campus when the knowledge ships to everyone at once. The customer stopped caring where you sit. What an enterprise buyer wants is a system that works, a team that owns it, and a price that makes sense. The postal code on the invoice is not on the list. Results travel. The talent was always here. India produces an enormous depth of engineering talent. What changed is that this talent can now build for a global market directly, as a product company, instead of only serving one as a back office. So we build for India and the world at the same time. INR pricing and local payment for Indian customers who were treated as an afterthought by foreign vendors. Global standards for everyone, because the bar is the bar. We are proud to be from Kolkata. Not as a discount. As a headquarters. The next wave of great AI companies will come from places the last wave ignored. We intend to be one of them. #IndiaTech #AIStartup #FounderJourney #Aptibit

Most VMS RFPs ask about resolution, storage, and camera counts. Those are easy for any vendor to answer and they predict almost nothing about how the deployment will actually go. Here are 12 questions that separate the platforms that scale from the ones that demo well. We would happily answer all of them. What is your concurrent connection ceiling per node and per deployment? What is your end to end latency for live view, measured, not marketed? Do you charge per camera, and how does cost grow as we expand? Which streaming protocols do you support natively, without a gateway? Where does our video physically live, and can it run fully on premise or air gapped? Which analytics are included versus a paid add on? Do your models adapt to our site, and what is the false positive rate after week one? How do you integrate with access control, intrusion, and fire, two way, not read only? What does the audit trail capture, and is it tamper evident? How does migration from our current VMS work, and can it run side by side? What languages and scripts does the operator interface support? Can you give a reference operating at our scale for 12 months or more? Notice what these have in common. They are hard to fake. A vendor either has a real answer or starts talking around the question, and the talking around is the signal. Steal this list. Put it in your next RFP. The responses will sort the field for you. #VMS #Procurement #RFP #PhysicalSecurity #Visylix

The fastest way to get bad software is to let the people who build it hand it off and walk away before they ever have to live with it. When the builder never runs the thing, every hard problem becomes someone else's. Corners get cut because the person cutting them will not be there at 2 am when it breaks. End-to-end ownership flips that incentive. The team that designs it, builds it, ships it, and runs it makes very different decisions, because they are the ones who get the call when it fails. What ownership changes in practice: The builder feels the operational cost. When you carry the pager for your own code, you write code that does not page you. Reliability stops being a checkbox and becomes self interest. The feedback loop closes. The person who sees the production problem is the person who can fix the design that caused it. No translation layer, no ticket thrown over a wall, no lost context. Accountability has a name. When something is wrong, there is no committee to hide in. One team owns the outcome, good or bad. That clarity is worth more than any process. Quality becomes personal. It is easy to ship mediocre work you will never see again. It is much harder to ship mediocre work you have to maintain for the next two years. This is why we do not split build and run at Aptibit. The team that ships your system is the team that stands behind it. Design, engineering, deployment, and the responsibility that comes after. Software gets good when the people making the decisions are the people who live with them. Own it end to end. It is the difference between a deliverable and a system you can trust. #EngineeringCulture #Ownership #AIDevelopment #Aptibit

In most industries, surveillance video is an operational tool. In banking, it is evidence. One day a regulator, an auditor, or a court will ask your footage to testify, and the system either holds up or it does not. That raises the bar from watching to proving. Visylix is built for the proving. What banking actually needs from a VMS: A complete, tamper evident audit trail. Who viewed what footage, when, and why. Every export logged. In a regulated environment, the record of who touched the evidence is part of the evidence. Video tied to transactions and access. The ATM event, the vault door, the teller station, correlated with the footage of the moment. When something is disputed, the timeline is already assembled. Data that never leaves the institution. Runs on premise, on the bank's own infrastructure, as a container. Customer footage does not travel to a vendor cloud, which is one fewer conversation with the regulator and the privacy office. Retention and legal hold done right. Policies enforced by the system, not by a person remembering. Footage held for a case is protected from the routine deletion schedule. Access control that matches the org chart. Not everyone should see every camera. Role based access, isolated by branch and function, auditable end to end. In banking, a VMS that merely records is a liability waiting for the day it cannot prove what happened. A VMS that records, correlates, and audits is an asset the same day. Build for the cross examination you hope never comes. That is the whole job. #Banking #Compliance #VMS #PhysicalSecurity #Visylix

We build products, not projects. It is on our wall, and it is not a slogan. It is a decision that changes every choice that follows. A project has an end date. A product has a life. Those are two different things to build, and the difference shows up in the code. What changes when you build for a product instead of a project: You optimize for the second year, not the launch. A project team races to the deadline and hands over a system nobody can extend. A product team builds so the thing can grow after they are gone. Clean architecture is not perfectionism, it is respect for the future. You measure outcomes, not deliverables. A project is done when the feature list is checked off. A product is done when it moves the number it was built to move. Those are rarely the same day. You treat the handover as a feature. A project ends in a document dump. A product ends with a team that can run it without you, because your goal was their capability, not your indispensability. You say no more often. A project takes every request to hit the scope. A product protects itself from the features that would bloat it into something worse. Restraint is a design tool. You stay after launch matters. A project team is gone at go live. Go live is where a product's real life begins, which is exactly when the project mindset walks out the door. Most software disappoints because it was built as a project and then asked to live as a product. The mismatch is the problem. Pick the word first. It decides everything after it. We build products. Even when the invoice says project. #ProductThinking #AIDevelopment #EngineeringCulture #Aptibit

Move fast or do it right. It is presented as a trade off. It is usually an excuse. Teams that are slow and sloppy exist. Teams that are fast and sloppy exist. And teams that are fast because they are disciplined exist. That last one is the one worth being. At Aptibit, production grade on day one is not a slogan. It is a set of habits that make speed and quality the same thing instead of opposites. Here is how the two actually reinforce each other. We ship a thin slice, not a big bang. A narrow path all the way to production, early. It is fast to build and it surfaces the real problems while they are still small. Big bang builds hide their problems until the end, which is the expensive place to find them. We write the tests as we go, not after. Tests are not the thing that slows you down. Debugging untested code in production is. Discipline up front is what lets you move fast later without fear. We automate the boring path. Build, deploy, check. If shipping is a manual ritual, shipping is rare and scary. If it is one command, it is routine, and routine is fast. We keep the code readable for the next person. Because the next person is often us, six months from now, and clarity is just speed you have not spent yet. Shortcuts feel fast on Tuesday. They are the reason Thursday is on fire. Real speed is not skipping the work. It is doing the work in an order where quality makes you faster instead of slower. Fast and good are not enemies. They come from the same discipline. #EngineeringCulture #AIDevelopment #Quality #Aptibit

A safety officer cannot watch 40 cameras at once, through the night shift, every night, forever. So the one moment the hard hat comes off, or a guard is removed, is usually the moment nobody is looking. That gap is where incidents live. Visylix closes it without a single new piece of hardware. On the cameras a manufacturing floor already runs: PPE compliance. Is the hard hat on. The high visibility vest. The safety glasses. The system watches every person in frame, all the time, and flags the exception the human eye would miss at 3 am. Smoke and fire detection. Visible smoke, opacity change, color shift in the frame, caught in seconds and cross checked against alarms, so a real event escalates fast and a false one does not scramble the whole plant. Restricted zone monitoring. A person in the path of a forklift. Someone in an area during a lockout. The alert reaches a supervisor in the moment, not the incident report. Zero downtime monitoring. It runs continuously, on premise, on your own infrastructure, so nothing sensitive leaves the plant and nothing waits on the cloud. The point is not to replace the safety officer. It is to give them eyes on every camera at once, so their attention lands where it is actually needed. The cameras were installed for security. On the factory floor, they are also a safety system nobody switched on. Switch it on. #Manufacturing #WorkplaceSafety #VMS #ComputerVision #Visylix

Most retrieval augmented generation systems fail for the same handful of reasons. Every one of them is decided before the first line of code. Here is the checklist we run at Aptibit before building a RAG pipeline. If these answers are shaky, the code will be too. What question is this actually answering? RAG is not a feature, it is a way to ground answers in your data. If you cannot state the question class it serves, retrieval has no target. Is the data clean, current, and permissioned? Garbage in is confidently wrong out. And if the model can retrieve a document the user should not see, you have built a data leak with a chat interface. How is the content chunked? Chunk too big and retrieval is noisy. Too small and it loses context. This unglamorous decision drives more quality than the model choice. What is retrieval quality on real queries? Before generation, measure retrieval alone. If the right document does not come back, no model can save the answer. Are answers cited and traceable? Every answer should point to its source. Citations are how users trust it and how you debug it. What happens when nothing relevant is found? The system must say I do not know, not invent. The refusal path is a feature, not an edge case. How does fresh content get in? A pipeline that is current on launch day and stale by month two is a maintenance problem you shipped on purpose. Notice the model barely comes up. RAG quality is a data and retrieval discipline. The model is the last mile, not the first. Get the checklist right and the build is straightforward. Skip it and you will debug in production instead. #RAG #LLM #AIEngineering #DataEngineering #Aptibit

Most retailers use their cameras for exactly one thing: watching for theft. The same cameras could answer the questions your merchandising team pays consultants to guess at. The camera is already there. The only thing missing is the software that asks better questions of it. What Visylix turns retail cameras into: Heatmaps. Where do customers actually go, and where do they never linger. The end cap you fought for might be a dead zone. The corner you ignore might be gold. Now you know instead of guess. Queue analytics. How long is the checkout line right now, and when does it spike. Staff the tills against reality instead of a schedule written last quarter. Dwell and flow. Which displays hold attention. How people move through the store. The path from door to purchase, measured. Loss prevention, smarter. Not just recording the theft for later, but flagging the pattern in the moment, with far fewer false alarms as the models learn your store. All of this runs on the cameras you already installed, on your own infrastructure, with no per camera licensing punishing you for the coverage you already paid for. The security team bought the cameras. The whole business can use them. Retail cameras are the most underused sensor network in the building. Point better software at them. #Retail #VideoAnalytics #VMS #LossPrevention #Visylix

The old outsourcing pitch was simple: cheaper. It also came with a quiet asterisk about quality, timezones, and rework. In 2026 that pitch has changed, and most Western buyers have not updated their mental model. The new math is not cheaper for less. It is the same quality for a fraction of the cost. That is a different conversation. Here is what actually changed. The talent gap closed. The senior engineers building AI in India trained on the same papers, the same open source, the same models as everyone else. The frontier is public. Geography stopped gating skill. The tools equalized. Cloud, models, and infrastructure cost the same in Kolkata as in California. The old advantage of proximity to compute is gone. The delivery model matured. A dedicated team that owns context beats a body shop that rents hours. The good firms sell outcomes and continuity, not headcount and timesheets. What you get when it is done right: Senior AI engineering at a fraction of tier one salary cost A dedicated team that keeps your context instead of resetting it Timezone overlap used as an advantage, not an excuse Clear ownership of code and data, on your terms The buyers still running the 2010 mental model are comparing on price and bracing for rework. The buyers running the 2026 model are getting global standard work and reinvesting the savings into shipping faster. Global standards. India pricing. That is not a discount. It is a structural advantage. #Outsourcing #AIDevelopment #IndiaTech #GlobalDelivery #Aptibit