Dipanjan Karanjai on architecting agentic systems that match their Day 1 design at Tredence
An Unlikely Starting Point
I did not start my journey in computer science. My undergraduate degree was in civil engineering, and my master's research at IIT Madras focused on submarine detection and vortex-induced vibrations for DRDO. Machine learning entered the picture not as a career plan but as the sharpest tool available for problems that classical methods couldn't crack. CNNs, LSTMs, RNNs — I was using them to model multi-physics problems before the current AI wave made them household terms.
That foundation shaped everything: I think about AI systems the way an engineer thinks about structures — load-bearing, failure modes, long-term integrity. Outside of work, I love playing chess, going for a run, and playing with my child.
The GOAT Moment
I cannot pick just one GOAT moment — there were three, and they're inseparable. First, standing in a client whiteboarding session as the primary technical voice in the room, pitching an org-wide agentic ecosystem architecture, and watching the client say: yes, exactly that. Second, delivering the project built precisely on that Day 1 design — no fundamental rethinks, no architectural retreats. Third, and most unexpectedly, earning the kind of trust where a client stakeholder just picks up the phone to share a new idea directly. That's not a delivery metric. That's a relationship, and it's the part that compounds.
We didn't just build to the original specifications. We continuously innovated and over-delivered, providing significantly more value than they had initially envisioned.
The GOAT Makers Behind Me
Ashish Jain pulled me into high-stakes client rooms early, which forced me to think at the systems level rather than the model level — a shift that permanently changed how I approach architecture. Bharadwaj Tadikonda is the person I stress-test ideas with; those conversations have sharpened more of my technical thinking than I can easily count. Sriram Gudimella kept pushing my scope outward — into CoCs, technical blogs, hiring panels — every time I was tempted to stay heads-down on delivery. And the broader Tredence team is the environment that makes all of it possible: a place where ideas actually get challenged, not just nodded at.
What I'm Building Towards
From the structural integrity of civil engineering to the high-stakes signal processing of towed-array sonar cables for submarine detection, my career has been defined by one challenge: extracting reliability from complexity. Today, I translate that expertise into Enterprise AI — building and deploying foundational models that solve complex problems within the "impossible" constraints of low-compute environments.
By architecting agentic AI and multi-agent workflows, I'm designing the next generation of autonomous systems — moving beyond static models to create resilient, self-orchestrating intelligence that can navigate the most intricate technical landscapes. Right now, that means continuing to push the frontier on agentic AI deployments, building the MLOps and LLMOps frameworks that keep these systems honest in production, not just impressive in a demo. The question I keep returning to isn't what AI can do. It's what it takes to make AI something an organisation can actually depend on.
Build Systems That Last. Join Tredence.
If you want to architect AI that solves real problems at enterprise scale — across BFSI, energy, shipping, and beyond — Tredence is where that work happens. We're hiring across Data Science and AI roles.