Conf42 Large Language Models (LLMs) 2026
2026-06-18
-- online --
[Conf42 Large Language Models (LLMs) 2025, Conf42 Large Language Models (LLMs) 2026]
| title | author | length |
|---|---|---|
| Knowledge Graphs as the Backbone of AI-First Software Delivery | Prateek Mittal | 14m |
Claims a production Knowledge Graph powers AI agents to query CI/CD, security, observability and ticket systems, using HQL queries over connected nodes… | ||
| Make Agents Remember the Right Context at the Right Time | Anannya Roy Chowdhury | 14m |
A game called Horcrux Hunt exposes memory failure modes and drives a concrete fix: a three-layer memory model (working, retrieval, persistent) plus off-model… | ||
| Teaching and LLM to Catch British Sarcasm | Anton Denisov | 16m |
11,000 landlord interactions from ~2,000 accounts feed a production pipeline that uses Vox transcription, Watson transcription for calls, BigQuery for… | ||
| Voice Agents That Think While They Talk | Bohdan Snisar | 22m |
Pipeline uses WebRTC 20-ms frames for realtime audio, FFmpeg for transport, and VAD via Silero or Deepgram; latency targets are under 500 ms latency budget… | ||
| Building Voice AI for People Who Don’t Trust AI | Pavel Shchekotov | 12m |
Built a phone-first voice agent for 55+ users that avoids apps, uses a natural female 11lab voice and is named Sophie, explicitly discloses it’s AI, and stores… | ||
| Reading is the New Writing in the AI Era | Petras Bazdaras | 07m |
Claim: developers spend 90% of time auditing LLM output because generated code compiles but hides anti patterns, deprecated APIs, and security gaps. The talk… | ||
| RAG for Data Engineers: Why Your Metadata Is the Missing Piece | Varun Joshi | 29m |
Argues that retrieval augmented generation (RAG) closes the gap between generic LLMs and proprietary warehouses by indexing existing metadata (DDL, DBT model… | ||
| Beyond the Context Window: "Mental Model" for Your AI Agents | Lior Schejter | 38m |
A searchable architecture hub is a plain Git repo of flat markdown files that an orchestrator populates by running parallel subagents to extract faceted… | ||
| LLM Agents as Co Pilots for Scalable Decision Systems | Mazdul Choudhury | 11m |
A three‑layer architecture claims to turn LLMs into reliable co‑pilots by combining a signal aggregation layer that unifies transactions, forecasts and… | ||
| The Judge Problem: Agentic Verifier is Rotting in Production | Vsevolod Voropaev | 27m |
Argues verifiers silently fail: second-model judges produce no alerts while emitting optimistic scores, driven by measurable biases like position bias and… | ||
| Building Real-Time AI Decision Systems with LLMs | Mohanraj Varatharaj | 24m |
Practical blueprint for running LLMs as real‑time decision engines: ingest events with Kafka, Kinesis, Flink, feed a RAG LLM reasoning layer, evaluate outputs… | ||
| LLM-Driven Financial Ledgers: Real-Time Accounting Systems at Scale | Rakesh Jha | 36m |
Claims a 30% drop in reconciliation time, 40% fewer manual interventions and a 35% faster month‑end close, backed by KPMG numbers, using real‑time event… | ||
| Automating Database Operations with Cortex Code | Timothy J Spann | 28m |
Automates Snowflake database DevOps end to end, generating SQL, DBT projects and pipelines from spreadsheets or Git, enforcing role-limited Snowflake account… | ||
| Shipping AI Features That Don’t Break in Production | Andrii Veremiienko | 23m |
AI cuts prototype time from ~6 months to ~2 months for the indie apps but introduces architecture drift, UI inconsistency, and debugging nightmares in both… | ||
| Open-Source Framework for AI-Powered CI/CD Pipelines | Denis Ermakov | 29m |
Claimed outcome: model-generated code is only ~10% of the work and the other 90% is structure, validation, cost controls and human gates. Built an open-source… | ||
| Spec Driven Modernization: Spec2Cloud & GHCP Modern CLI | Marco Antonio Silva | 48m |
Shows a spec-driven pipeline that converts a legacy .NET Framework 4.8 MVC app into a running .NET 10 service using deterministic and agentic tools, with… | ||
| Diffusion to Prod: vLLM-Omni Unifies Multimodal Inference | Kosseila Haddalene | 45m |
A single runtime unifies autoregressive and diffusion generation so one pipeline handles text, image, audio and video rather than two stacks, implemented as… | ||
| Agentic AI for Autonomous Compliance in Enterprise Tax System | Abdul Basit Iqbal | 22m |
Claims a multi-agent Autonomous Compliance Infrastructure compresses a 3–6 month regulatory-to-production cycle into hours by combining continuous regulatory… | ||
| From Acceleration to Assurance | Karishma Velisetty | 18m |
Nine concrete SOX use cases map to specific controls and risks with tools like GitHub Copilot, Cursor, Claude, ChatGPT used for code review and script… | ||
| Validation Pipeline for Summary Evaluation | Liliya Imasheva, Michael Banf | 14m |
Built a reproducible LLM-as-judge pipeline that scores summaries on five expert-defined Likert dimensions, binarizes them (1–3 = bad, 4–5 = good), and reports… | ||
| Compliance-by-Construction for Credit Apps | Ravindra Rajasekhar Kavuru | 15m |
Claims an architecture that enforces regulatory invariants at design time and proves it with measurable outcomes: targets under 1 ms end-to-end latency… | ||
| AI Product Management for Production LLM Lifecycles | Vijayalakshmi Narasimhan | 22m |
Argues that production LLMs need concrete product requirements: require explainable responses with cited sources, access controls and end-to-end audit logs for… | ||
| Log-Centric Intelligence: Enabling Real-Time Compliance O11y | Prem Anand | 32m |
Argues that treating structured application logs as the primary intelligence layer — via a wrapper library that auto-instruments services, injects correlation… | ||
| Agile Transformation for Scalable Healthcare IT | ShubhaPriya Muthuramalingam | 15m |
Healthcare IT shifts from years to weeks by adopting two week sprint cadence, embedding HIPAA controls, access management, audit logging and data lineage as… | ||
| LLM Enabled Multi Document Correlation for Financial Compliance | Varsha Shah | 31m |
Claims a working pipeline that ingests ~3 million records across four jurisdictions and five years and combines a graph-based entity correlation engine, an… | ||
| ML Automoderation System for LLM-Augmented Online Platforms | Andrei Shcherbinin | 30m |
Production pipeline moves raw chats through cleaning, deduplication, sampling, hash journal checks and few‑shot retrieval, then uses LLM pre-annotation as a… | ||
| Shift-Left Performance Engineering for RAG and LLM Platforms | Kandasamy Selvaraj | 38m |
Shift-left CI/CD adds an automated deviation score gate that evaluates time-to-first-token, token count, LLM call count, tool latency and error/retry rates… | ||
| Clean and Controlled AI with API Management | Leo Visser | 30m |
Demo shows an Azure AI Gateway (API Management) fronting Microsoft Foundry and GPT-5 mini, enforcing a 60 tokens per minute per‑subscription limit (demo… | ||
| Building Reliable Infrastructure for LLM Workloads on K8s | Sergey Speranskiy | 13m |
Claims a production-ready K8s pattern: use a BGP anycast VIP for API server failover, peer each node with two top‑of‑rack switches and run bird on the host for… | ||
| Building Reliable LLM Systems | Lakshmi Priya Gopalsamy | 20m |
Models stall not from performance but from human queues: the talk shows decision latency rising from seconds to days, recommends a three‑tier gate… | ||
| LLM Copilots for SRE: Incident Response & Auto-Remediation | Susanta Kumar Sahoo | 25m |
LLM copilots claim to reduce analysis time by about half by ingesting logs, metrics and traces from Grafana/DataDog/Splunk, correlating deployments and… | ||
| Identity-Centric Security for AI Systems | Arunkumar Muthuramalingam | 16m |
Argues identity becomes the primary control plane for AI, not a user admin task, and lists ten common failure patterns; non-human identities now outnumber… | ||
| Your AI Has Too Much Access: Authorization for Coding Agents | Sohan Maheshwar | 30m |
Claims that coding agents need deterministic, fine-grained authorization and shows an open source stack that implements it: SpiceDB (Zanzibar-style graph… | ||
| RuntimeGuard-AI: Cryptographic Compliance | Neeraj Kumar Singh Beshane | 25m |
RuntimeGuard-AI produces cryptographic verifiable receipts for every consequential agent decision: signed digests of prompt, retrieval context, policy version… | ||
| nothing matches | ||