Sanjay Sharma

🚀 From CI Failures to Autonomous Healing — Our RIFT 2026 Build At RIFT 2026 (organized by PW (PhysicsWallah), we challenged ourselves to rethink how CI/CD pipelines work. Every developer knows this loop: Push → CI fails → Debug → Fix → Push again → Repeat. What looks like a small syntax error often costs hours of context switching and delay. So instead of asking “How do we improve debugging?” We asked: What if CI didn’t just fail… but healed itself? 🧠 What We Built We built an Autonomous CI/CD Healing Agent with a React Dashboard — a system that: • Takes a GitHub repository as input • Automatically discovers and runs tests • Parses logs and classifies failures • Generates minimal targeted fixes • Creates a deterministic branch • Commits with [AI-AGENT] prefix • Pushes and re-triggers CI • Iteratively retries until convergence All execution runs inside a Docker sandbox, with a configurable retry limit and strict safety controls. This isn’t just automation. It’s structured, bounded autonomy. 🔗Website Link- https://lnkd.in/dYrenM2f 🏗 How It Works We designed it in layers: Frontend (React Dashboard) Collects inputs and renders structured results via results.json. Backend Orchestrator Manages execution flow, retry logic, and safety boundaries. Multi-Agent System • Repository Analyzer • Test Execution Agent • Failure Interpreter • Fix Generator • Git Automation Agent • CI Monitor Agent Each agent has a single responsibility, making the system modular, scalable, and explainable. At the core lies a closed-loop healing system: Run → Analyze → Fix → Commit → Re-run → Verify → Repeat. That feedback loop is the real innovation. 🔐 Why This Matters CI failures aren’t just technical issues — they’re productivity bottlenecks. Developers spend 40–60% of their time debugging pipeline failures. We’re not building a debugging tool. We’re redesigning the CI/CD feedback loop. 📊 Built with Engineering Discipline We implemented: • Structured log parsing • Controlled retry logic (max 5 iterations) • Deterministic branch creation • Safe commit automation • CI status monitoring • Score calculation logic • Strict test-case matching Built end-to-end — frontend, backend, orchestration, GitHub API integration, sandbox execution, deployment — under hackathon pressure. And this is not the end. 💡 We’re already planning: • Slack integration • GitHub App plugin • Enterprise-grade version • Self-hosted runner • Smarter prioritization • Per-repository pricing and many more... This isn’t just a student project. It’s a scalable DevOps automation layer. 🔗Website Link- https://lnkd.in/dYrenM2f Proud of what we built. Big thanks to Kushagra Garg Excited for what’s next 🚀 #RIFT2026 #PhysicsWallah #DevOps #AI #SystemDesign #AgenticAI #CI_CD #Hackathon #SoftwareEngineering

23

1 Comment

RAG enhances LLM responses by grounding them in your enterprise’s data. Once the knowledge base is embedded into vectors, the query pipeline takes over— Converts questions into vector form, performing semantic similarity search, and injects relevant info into the prompt for accurate, context-rich answers.    Powered by SAP HANA Cloud’s Vector Engine and the Generative AI Hub, this pipeline supports semantic retrieval, integrates external domain-specific data via prompt templating, and safeguards sensitive information through data masking.    This approach leverages foundational techniques like vector similarity search to surface semantically aligned content, chunking strategies to optimize the context window, and hybrid search to combine the nuance of vector semantics with the precision of keyword filters — boosting accuracy, adaptability, and relevance.    Key benefits: ➡️ Dynamic Adaptability – Handles unstructured input with accuracy.  ➡️ Reduced Hallucinations – Matched context reduces irrelevant outputs.  ➡️ Contextual Relevance – Answers aligned with meaning, not just keywords. RAG pipelines go beyond smart—they’re strategic, transforming how enterprises retrieve, interpret, and generate knowledge with real-world context and reliability.    Learn more: https://lnkd.in/eifTsc_D

146

4 Comments

How LLM Inference Works by Arpit Bhayani #LLM #inference transforms text prompts into responses through a process involving #tokenization, #transformer layers with self-attention mechanisms, and autoregressive #tokengeneration. There are two stages in practice: the model first handles your full prompt in parallel, then switches to generating tokens one by one, which shifts the bottleneck from math to memory access. Key optimizations include KV caching to avoid redundant computation, batching to improve GPU utilization, and quantization to reduce memory pressure. https://lnkd.in/eJkBiyWE

1

To add to that, Confluent Cloud Network (CCN) routing simplifies private networking for Flink, reusing existing Kafka infrastructure. It helps teams quickly connect secure pipelines across cloud environments. Security is further strengthened with IP Filtering—giving teams greater control over access in hybrid environments and enabling them to confidently deploy AI at scale. With these enhancements, Confluent is not just enabling smarter AI—it’s making it enterprise-ready. Organizations can now unlock: Seamless integration of historical and streaming data Secure, scalable infrastructure for AI agents Faster, context-aware decisions powered by trusted data Agentic AI is here. With Confluent Cloud, it’s ready for the enterprise. https://lnkd.in/gh-dA7qs #ConfluentCloud #ApacheFlink #SnapshotQueries #AgenticAI #RealTimeData #StreamProcessing #DataAnalytics #EnterpriseAI #BatchProcessing #DataStreaming #DigitalTransformation #Flink #AI #MachineLearning #DataSecurity #PrivateNetworking #IPFiltering Follow our LinkedIn page for more such interesting CX expert opinions/ Thought Leader Interviews/ CX Leadership news/ articles/ updates https://lnkd.in/g9eeFFgh

1

37% ARR growth for an MDM vendor is truly remarkable. That's because the market itself usually only grows around 6-9%, which means that Profisee is clearly doing something right. The success of Profisee speaks to the necessity of MDM in a modern data stack - but more importantly - it speaks to the product vision, the unwavering customer focus, and the commitment of the entire Profisee team. I'm humbled to be a part of this company, and I'm excited to see where things head. Something tells me we're just getting started... #mdm #masterdatamanagement #masterdata

18

1 Comment

If you need a custom LLM and have massive domain data, JAX offers advantages over PyTorch for training—especially for distributed/TPU workloads & with functional programming. This nice course breaks down JAX for LLM development - https://lnkd.in/gh8phAtB

32

great post on the intersection of AI workloads and Open Container Initiative "The State of OCI Artifacts for AI/ML" https://lnkd.in/gbXZykMq

47

The importance of metadata and data semantics is back in the spotlight with the recent OSI – Open Semantic Interchange. https://lnkd.in/dXCEFucy Just like the old “plug and pray” boards revealed configuration struggles, today’s AI agents will only succeed if they rely on consistent semantic models — something often as hard as getting developers to value the importance of semantic models, and managers to appreciate the relevance of modern data persistence. These models strengthen data governance, requiring organizations to describe their data right at the source. Their true power emerges when connected to normalization and interoperability layers such as MCP (Model Context Protocol), or when linked to rules expressed as code in critical domains like tax compliance. The adoption of any standard largely depends on its market penetration — just as we’ve seen with USB-C (no more PnP boards), which became universal thanks to its interoperability power. Industry and regulators must also play their role by promoting adoption, setting clear requirements, and ensuring interoperability frameworks become the backbone of trust and compliance. At SOVOS, we are designing our products with these trends in mind, staying close to our clients, and ensuring their Business/IT alignment.

18

In this new world of being “drunk on compute”, we always want to throw more compute to make a bad query run faster. But what if we paused and addressed the problem of query performance? See my new blog on How to address query performance challenges in Snowflake : https://lnkd.in/eN3HCZFV

19