ThirdEye Data

10/16/2026

We work as a team and enjoy as a family. In September, the family went to Kuzhupilly Beach in Kochi. 🌴

Team members from across India met at Bengaluru airport. Some arrived with neatly packed bags, and some had clearly been packing until 3 am. 😂
Our Founder and CEO, Debjyoti Das, flew all the way from the US to join the India team, which gave him the best jet lag excuse in the group. 😉

We stayed at 36 Palms resort near Kuzhupilly Beach. Within an hour of check-in, the laptops were forgotten, and everyone was on the shore trying to find their good side for the golden hour photos. We took group shots, candid shots, and roughly forty retakes, because somebody always blinked. 📷📷

The backwater boat ride was the calmest any of us had been all year. There were no Teams pings and no client calls, just green water, mangroves, and a boat full of colleagues who suddenly became nature photographers and selfie geniuses. 🤳

At the Annual Gala, Dj walked us through the year and recognised the people who made it happen. Then the music started. We learned that some of our engineers debug code much faster than they dance, while a few of them dance far better than anyone expected. The prestigious “ThirdEye Data Annual Boat Dancer” award goes to Shahi, Mukherjee, Sai Pratyusha and Akalwadi. 🕺💃

In Fort Kochi, we visited St. Francis Church, which was built in 1503. That makes even our oldest legacy systems look young. We wandered through the old lanes of Princess Street on purpose, and a few of us also got lost in them by accident.

We came to Kerala for the beaches and stayed for the fish curry and appams. Nobody is admitting how many calories they had at the farewell lunch.

We came back with full camera rolls, slightly fuller stomachs, and the same belief we left with. The team that solves hard AI and data problems together on Monday is also the team that cannot stop laughing together on Saturday. That is the culture we want to keep building at ThirdEye Data .

10/14/2026

Job Alert! We're hiring Sr. AI Engineers.

At ThirdEye Data, we are looking for experienced Senior AI Engineers with 7+ years of professional experience to join our engineering team.

The candidates will be responsible for designing, developing, deploying, and optimizing enterprise-grade AI and Machine Learning solutions using state-of-the-art Microsoft tech stacks and cloud platforms.

Check the complete job details here and apply:
https://buff.ly/UNegRxU

10/12/2026

“What have you actually built for warehouse teams?”
That was one of the most common questions visitors asked on our website over the last few weeks, so here is a straight answer.

At ThirdEye Data, we have AI solutions running in production across four areas of warehouse operations:
👉 Inventory counting and quality inspection
👉 Space utilization
👉 Workforce safety and compliance monitoring
👉 Supplier operations workflow automation

Inventory counting and quality inspection is the one customers ask for most, and it delivers the clearest return of the four. More than seven customers already run it.

Here is what it looks like at one of them. A commodities company runs stock counts across more than 2,200 warehouses with an app on the smartphones its staff already carry. A staff member photographs the stack, our vision model counts the bags, and the count is saved with a timestamp, so head office sees the same number the warehouse sees.

Counting accuracy is above 93%, and the company installed no new cameras or scanners. For everyday stock counts, the model runs on standard CPUs, which keeps the cost per site low enough to roll out to every location. When a site needs faster counting, such as live counts from floor camera video or thousands of photos at month-end, we add a GPU. A GPU processes many image frames in parallel, so counts come back in seconds even at high volume.

If you need to automate stock counts, inspect quality, monitor safety compliance, or make better use of your space, we can have a working pilot running on your own site in three to four weeks. Book a 15-minute walkthrough; here is the link to explore more details:
https://buff.ly/t1JgQLM

10/08/2026

Nearly half of companies now get real value from AI, up from about 5% a year ago, according to BCG's 2026 Applied AI Index.

The interesting part is where that value is coming from. Here are the agentic AI use cases delivering for enterprises in 2026:

- Customer service: HubSpot's AI agents now resolve more than 70% of its support tickets. One airline's voice agent handled 6 million calls in its first year and saved $8 million.

- IT service desk: Tech Mahindra reduced first-level IT support work by about 25% across 150,000 employees using agents on ServiceNow.

- Finance back office: Gartner found that invoice processing, receivables automation, and report creation pay back within 9 to 10 months, faster than forecasting or open-ended analysis.

- Supply chain planning: One Fortune 500 industrial company cut its demand-to-production planning cycle from 30 days to 3, with about 80% of planning now touchless.

- Sales: Siemens now responds to every one of its 2,500+ monthly inbound leads across 132 countries with sales agents, where many used to go unanswered.

- Marketing: Klarna cut campaign production from six weeks to seven days while producing more content.

- HR: Standard Chartered reduced the onboarding effort for hiring managers by 35%.

None of these is a moonshot. Each one takes a workflow that runs at high volume and already has clear rules, lets the agent finish the routine cases, and keeps a person on the exceptions. That pattern is the most reliable signal of where agentic AI will pay back next.

We mapped 30 agentic AI use cases that fit it, by function and industry, with the value metrics to track for each one. Read the full article here:
https://buff.ly/mU6rYIb

Which of these workflows is your organization prioritizing for 2027?

Top 30 Agentic AI Use Cases in 2026 Agentic AI solutions are delivering

10/05/2026

Happy World Architecture Day to all the architects, designers, and drafters out there. 🏛️

ThirdEye Data is a data and AI engineering company, so this might seem like an odd day for us to post about. But a project we worked on changed how we look at your work, and today felt like the right time to share it.

Ask anyone who has worked on a renovation or fit-out what the first week looks like. The client sends over existing conditions, and it's rarely a clean DWG. More often it's a scanned as-built from years ago, covered in redlines, with door tags you can barely make out. Somebody has to redraft the whole plan in CAD before design work can start, then go through it again to put together the door and window schedules.

The design teams we worked with were losing a lot of hours to exactly that. So we built an AI solution that reads floor plan images, finds the doors, windows, and fixtures along with their tags, and turns it all into a DXF file that opens in AutoCAD or ArchiCAD.

Honestly, the hardest part for our engineers was learning to read plans properly. Door swings, wall thicknesses, symbols that every office draws a little differently. We came out of it with a lot more respect for what goes into a single sheet.

Architects shape the places we live and work in. If we can take some of the redrafting off their plate, that's a contribution we're glad to have made.

Happy World Architecture Day!

09/29/2026

Every fresher today has the same question: "How do I actually learn to use AI, not just talk about it?"

That's exactly why we built Neev (नींव), part of Learning Path, ThirdEye Data's initiative to help students and young professionals build real, applied AI skills.

You won't sit through slide decks or theory you'll forget by next week. You're building from day one.

Self-paced, hands-on learning in:
➡️ How LLMs actually work
➡️ How RAG pipelines are built
➡️ Prompt engineering that gets you real output
➡️ Building AI agents that automate real workflows

You apply all of it to your own domain. Mechanical, Civil, HR, Marketing, CS, whatever field you're in.

Several courses are free to start. No financial barrier to begin.

Want in? Comment "Learn" below, and we'll share the details.

09/28/2026

Most AI solutions for document-heavy operations stops at extraction. Or too expensive. But we have a solution now!

Think about the scenario after data extraction with OCR:
⚠️ Someone pulls files from the drive.
⚠️Someone validates each one against the criteria.
⚠️Someone keys the data into the ERP.
⚠️ Someone builds the weekly report.
⚠️Someone digs through hundreds of documents to answer one question.

That is where the hours go.

Optira automates all of it.

→ Picks up new files from OneDrive, Google Drive, Azure or a local folder. Automatically.

→ Reads PDFs, Excel, CSV, scans, handwritten notes, and engineering drawings. OCR, computer vision, custom ML models and LLMs work together on every file.

→ Validates every document against your rules. You write them in plain language. An AI agent applies them.

→ Sends clean data straight into your ERP, CRM or MES.

→ Builds and sends reports by email or WhatsApp, hourly or daily.

→ Answers questions in chat. Ask "Which invoices failed validation this week?" and get the list.

Your team stays in charge. They review exceptions, not every page.

Your systems stay. Optira works on top of what you already run.

Your data stays. In your cloud or on your servers. Not with us. Not with any AI model.

the full chain, or start with the step that costs you the most hours. Optira delivered field-level accuracy up to 99.6%.

10,000 documents a month or 100,000. Pricing follows your volume.

Send us your hardest documents. We'll run them live in a demo.
Explore here: https://optira.ai/

Which of those five tasks takes your team the longest?

09/24/2026

Most enterprise AI pilots in do not fail in the demo.

They fail in week six of production, when someone opens the audit trail and finds the agent closed a P2 incident it never actually diagnosed. It guessed. The guess matched a runbook. Nobody caught it until the incident came back a third time.

We hit that on an agentic for project. The fix was not a bigger model. It was one arrow.

A standard LLM deployment runs in one direction. Ticket in, classification out. There is no step where the system asks whether its own assumption is true.

We added a return path. The agent forms a hypothesis, tests it with a read-only probe against live telemetry, and compares the result against what it expected. When the two do not match, it prunes that hypothesis and tries the next one. It is allowed to be wrong in a way that costs nothing.

That pattern has a name in neuroscience. .

For the engineers, the interesting part is the probe. Keep it as deterministic
API calls, not model-generated code. If the probe is itself probabilistic, you have not added verification; you have added a second guess.

For anyone signing the budget: around 70% of inbound tickets now can be closed with no human involvement. on escalations dropped 62%. Median triage went from 45 minutes to 12 seconds.

And the control that made it approvable was not accuracy. It was a ceiling. High-consequence actions have no autonomous code path at all. Not a threshold set very high. An absent code path.

The full architecture breakdown, four stages, and the risk tiers are shared in our latest article. Read here:
https://buff.ly/ON2kdQw

How We Use Neural Decision-Making Feedback Loops to Automate ITSM Ticket Triage Most enterprise AI pilots in IT service management

09/09/2026

Meta launched Muse yesterday. A personal that reads your email, touches your payment info, and books your trips.

The headlines weren't about what it can do. They were about whether you should let it.

"Can you trust it with your email and payments?" That's the story that stuck.

We are not here to review Muse. We build AI agents like this for a living, so here's why that question matters more than the product itself.

That's the exact question every enterprise runs into the moment an agent moves from demo to production. Not "is the model smart enough?" It's: what is this thing allowed to touch, what happens when it's wrong, and can you prove what it did after the fact?

Meta is asking millions of consumers to take that on trust, all at once.
Enterprises don't get that luxury. We've built systems for many support and operations workloads, and the pattern holds every time: the model is rarely the hard part. Scoping access, logging every action, building in human checkpoints, and proving the thing actually reduces work- that's the real build.

is a signal, not a template. Autonomous agents touching real data and taking real actions just went mainstream. Your customers will expect this next. The question is whether your agent's trust and governance are designed in from day one, or explained after something breaks.

Curious how others here are thinking about the access and audit side of agent rollouts, not just the model side.

09/02/2026

Learning AI shouldn't mean pausing your job, your semester, or your life for months.

Neev (नींव) is self-paced. Go as fast or as slow as your week allows.

But self-paced doesn't mean unsupported. Instructors are right there with you, guiding you through hands-on projects, so you're not stuck rewatching a video hoping it clicks the second time.

And every project is tied to your domain. Not "AI in general," but AI applied to the everyday work you actually do, whether that's Marketing, HR, Civil, Mechanical, or CS.

The registration window is open now. Spots are limited.
Comment "Learn" and we'll send you the details.