DevDay Pune

Software Engineering in the Age of AI

Sat, Oct 24 · 10:00 IST

Business Bay, Golf Course Square, Airport Rd, Jayprakash Nagar, Yerawada, Pune, Maharashtra 411006, India, Pune, MH, IN

🎟️ Please RSVP here to join-in for this conversation in-person.

Software Engineering in the Age of AI

Topic 1: A Green Harness Is Not a Healthy Harness
Instrumenting your coding agent’s harness against silent failure

Most teams that build a coding-agent harness (a CLAUDE .md, a few hooks, a CI gate) treat it as config they set once. It isn’t. The harness is supposed to guard your code changes. Those same changes erode the harness itself: the standing-rules file goes stale until it’s confidently wrong rather than usefully absent, rule density crosses an adherence cliff where even frontier models stop following it, and the agent doing the work can satisfy a failing gate by deleting the test that enforces it. Nobody’s watching the watcher, and a harness that fails silently is worse than no harness at all.

This talk gives you a failure taxonomy for the harness itself and the safety net for each failure. The core distinction: encoded guardrails (lint, tests, static analysis, anything the agent doing the work can edit or delete) versus structural guardrails (permission hooks the agent’s own session can’t switch off, default-deny network egress, a CI gate with no admin-bypass path). The first kind only holds if the agent cooperates; the second holds regardless of what the agent tries. “Coverage is theater” generalizes: every encoded guardrail is theater against a suggestible actor.

From there, three concrete practices for treating the harness as production infrastructure rather than settings: config-drift detection in CI, canary load-lines that confirm context actually loaded, and harness regression evals that gate every change to a prompt, tool description, or iteration cap before it ships.

Pre-requisites for attendees
Engineers and teams running coding agents against real production codebases who’ve already written a CLAUDE.md/AGENTS.md or a hook and want to know whether it’s actually holding. Includes people building agent infrastructure, harness tooling, and evaluation frameworks.
ㅤ

Topic 2: The Hourglass-Shaped SDLC
Most software teams still organise delivery around implementation. Estimates are often driven by the expected complexity of the implementation, progress is measured through the development stage, and testing and review wait for the implementation to finish.

That arrangement made sense. Implementation consumed most of the effort needed to turn an idea into working software. Building also forced the design to meet the codebase and change under pressure.

AI changes the balance by reducing the human attention spent on implementation. Work around implementation continues to consume substantial human attention. Teams still define the problem and choose a direction before code is written. After code exists, somebody has to decide whether the result is correct, useful, and safe to release.

If the width of each stage represents its share of human attention, the SDLC starts to look like an hourglass. Adding a faster coding tool to the old workflow accelerates the part already taking less human time, while decisions and evidence continue to set the pace for software delivery.

Software output can then rise faster than confidence, which is why this talk argues for an SDLC organised around decisions before implementation, evidence after implementation, and learning that carries production feedback into the next change.

Meet the Speakers
Karun Japhet
Solution Consultant, Sahaj Software
Karun is an engineer and consultant at Sahaj Software. I work on large-scale systems, architecture, and AI in software development. Most of my time goes into helping teams design systems that hold up as they grow especially as AI becomes part of how we build.

Sujit Kamthe
Solution Consultant, Sahaj Software
Sujit is an engineering leader, software architect and AI practitioner with more than 16 years of experience building distributed systems, cloud platforms and enterprise applications, and building high performing engineering teams. His current work focuses on production-grade agentic AI, cloud-native architecture and AI-assisted software engineering. He writes and speaks about software architecture, engineering practices, developer productivity and the changing role of engineers.

Agenda

• 10:00 AM – 10:15 AM: Entry & Networking
• 10:15 AM – 11:15 AM: Talk 1 – A Green Harness Is Not a Healthy Harness
• 11:15 AM – 11:30 AM: Coffee Break & Networking
• 11:30 AM – 12:30 PM: Talk 2 – The Hourglass-Shaped SDLC
• 12:30 PM onwards: Networking

Entry to the event will close at 10:30 AM.

🎟️ Please RSVP here to join-in for this conversation in-person.

Agentic AI in Production

Sat, Nov 21 · 09:30 IST

Koregaon Park, Koregaon Park, Pune, Maharashtra 411001, India, Pune, MH, IN

Note
Important Registration Information
This is an invite-only event with limited capacity.
To be considered for attendance, please register using the registration link provided in this Meetup description. Submitting an RSVP on Meetup alone will not be considered a valid registration.
All registrations will go through a screening process.
If your registration is approved, you will receive a confirmation email.
Link for registration- https://sahaj.ai/events/agentic-ai-in-production/

About the Event
Agentic AI in Production is a one-day conference bringing together practitioners who have built, tested and evolved agentic systems beyond the prototype stage.

Through real-world systems and hard-earned lessons, the sessions explore architecture, multi-agent coordination, reliability, evaluation, memory, observability and performance.
The focus is not on what agents could do, but on the decisions, failures and patterns that determine whether they actually work in production.

Sessions
Shift-left Considerations in Agentic AI
Karthika Vijayan
Agentic system design starts with getting the mental model and problem scope right, rather than jumping straight into agents and orchestration. This talk presents a shift-left approach to thinking about agentic systems, with capability, accuracy and consistency, performance, and modularity considered from the outset. The goal is to build the right system for the problem while anticipating the constraints that will shape it as it grows.

Engineering Agents Beyond the Happy Path
Shruti Dhavalikar
Building an agentic solution for a large enterprise platform comes with a unique requirement: building not one assistant, but a reusable framework that supports multiple use cases. This talk walks through the design and evolution of that system from three angles: performance, configurability and reusability, and robustness. It explores how a working-but-slow orchestrator can be rebuilt, how modularity enables reuse across use cases, and what evaluations can and cannot catch.

My Agents Can’t Agree and Now It’s an Architectural Problem
Yahya Poonawala
As natural language semantic complexity increases for an application, building reliable agentic systems becomes more challenging. This talk explores the evolution of a digital advertisement campaign planning assistant from a nominal orchestrator-worker system to an NLP-incorporated, consistency-assured multi-agent system.

Lessons from an Agentic AI Raga-Metal Band
Sujit Kamthe
This talk explores how eleven specialist agents collaborated to compose raga-metal fusion music using advanced agentic patterns for coordination, critique, and verification. Through shared-blackboard collaboration, evaluator-optimizer loops, deterministic verification, LLM-as-a-judge, and bounded multi-agent debate, the talk examines what belongs in code, what requires model judgement, and how agents should be decomposed around decisions.

Who Watches the Agent? Engineering Trust into Autonomous AI
Suman Paul Choudhury
Building an AI agent is easy. Trusting one in production is not. This talk explores the harness needed to bridge that gap, from evaluating multi-step agent behavior and enforcing guardrails to making agents observable, debuggable, and reliable. Drawing on the latest research in agent reliability, it looks at practical systems and techniques for turning autonomous agents from unpredictable prototypes into trustworthy production systems.

Lightning Talk: Principles & Strategies for Agent Memory
Amit Bhagat
Building a basic agent is easy; giving it reliable memory is hard. As interactions grow, naive state management can lead to surging token costs, latency spikes, and context blindness. This talk focuses on the principles of agent state and memory management, and evaluates different strategies for building reliable agent memory.

Lightning Talk: How Hidden Prompt Breaks Production LLMs
Vinayak Kadam
When LLM-powered applications break in production, engineering teams usually audit system prompts or blame upstream model updates. But system prompts are only half the instruction. Dynamic annotations, schema descriptions, and few-shot payloads injected at runtime can carry equal weight. This talk uncovers hidden failure modes in data-prompt coupling, examines common anti-patterns, and covers practical strategies to retain alignment.

Meet our Speakers
Karthika Vijayan
Solution Consultant
Karthika Vijayan has been conducting research in the field of conversational AI with voice and text data for almost a decade. Her research has been published in several journals and presented at various international conferences. Prior to joining Sahaj, Dr. Vijayan worked as a research fellow in the Department of Electrical and Computer Engineering at the National University of Singapore. She also worked as research associate in the Department of Electrical Engineering at the Indian Institute of Science, Bangalore. She obtained her PhD in Speech Signal Processing from the Department of Electrical Engineering at the Indian Institute of Technology Hyderabad in 2016. She is an active member of IEEE, ACM and APSIPA.

Shruti Prasanna Dhavalikar
Solution Consultant
Shruti Dhavalikar is a skilled Data Scientist with over 7+ years of experience in the field. With a deep passion for data science and a strong understanding of data patterns, she successfully delivers end-to-end product cycles as well as actively contributes to research projects. She has published and presented research works at several international conferences.

Yahya Poonawala
Solution Consultant
Yahya Poonawala is a technology professional with over 13 years of experience in software engineering, architecture and engineering leadership. Having worked across product and services organisations, Yahya brings a problem-driven approach to system design, with a keen interest in building scalable, reliable and reusable software systems. His current work at Sahaj focuses on applying these principles to complex engineering challenges, including agentic AI systems.

Sujit Kamthe
Solution Consultant
Sujit is an engineering leader, software architect and AI practitioner with more than 16 years of experience building distributed systems, cloud platforms and enterprise applications, and building high performing engineering teams. His current work focuses on production-grade agentic AI, cloud-native architecture and AI-assisted software engineering. He writes and speaks about software architecture, engineering practices, developer productivity and the changing role of engineers.

Suman Paul Choudhury
Solution Consultant
Suman Paul Choudhury is a Data Science Consultant at Sahaj with over 10+ years of industry experience in AI and Machine Learning. His work spans Generative AI, Computer Vision, AdTech, customer analytics and large scale ML systems.Suman has extensive experience working with technologies such as PySpark, deep learning, graph based machine learning, LLMs, LangGraph and modern MLOps ecosystems.He holds a Master’s degree in Signal Processing, has contributed to IEEE research publications and has patent involving vision assisted checkout systems.Suman is particularly passionate about translating complex AI concepts into practical systems that solve real business problems and about helping teams move from AI experimentation to scalable, production ready solutions.

Amit Bhagat
Solution Consultant
Software Engineer at Sahaj with experience spanning high-throughput telemetry platforms, agentic AI systems, and distributed data engineering. Brings a strong foundation in building resilient backend systems, real-time event-streaming architectures, and cloud infrastructure.

Vinayak Kadam
Solution Consultant
Vinayak Kadam is a Solution Consultant at Sahaj Software, where he works on AI-powered and multi-agent systems. With experience spanning data engineering, cloud platforms, and large-scale distributed systems, he focuses on building production-ready solutions that combine intelligent automation with practical business outcomes.

Agenda
9:00 AM – 9:30 AM
Registration and Networking

9:30 AM – 9:40 AM
Introduction

9:40 AM – 10:20 AM
Shift-left Considerations in Agentic AI
Karthika Vijayan

10:25 AM – 11:05 AM
Engineering Agents Beyond the Happy Path
Shruti Dhavalikar

11:05 AM – 11:20 AM
Break

11:25 AM – 12:10 PM
My Agents Can’t Agree and Now It’s an Architectural Problem
Yahya Poonawala

12:15 PM – 1:00 PM
Lessons from an Agentic AI Raga-Metal Band
Sujit Kamthe

1:00 PM – 2:00 PM
Lunch Break

2:05 PM – 2:45 PM
Who Watches the Agent? Engineering Trust into Autonomous AI
Suman Paul Choudhury

3:35 PM – 3:50 PM
Lightning Talk: Principles & Strategies for Agent Memory
Amit Bhagat

3:55 PM – 4:10 PM
Lightning Talk: How Hidden Prompt Breaks Production LLMs
Vinayak Kadam

4:10 PM – 4:25 PM
Break

4:25 PM – 5:30 PM
Birds Of A Feather / Open Discussion