Aman Kumar

@elvan_hq This thread is gold. The way you broke down why email surveys only catch promoters and miss the quiet churners is eye-opening. That in-the-moment trigger point makes total sense now. Really well explained. Thanks for sharing this.

Elvan was built for exactly that kind of feedback. In-product surveys that hit the right people at the right time. elvan.ai #NPS #CustomerFeedback #CX #SaaS #BuildInPublic #Founders #ProductHunt #Zendesk

Most feedback programs survey the wrong people.

@JainMayank237 Upvoted

Every product team today is collecting feedback… but very few are actually *using it to drive decisions. ⏰ Feedback comes in too late 🧩 It’s scattered across tools 📄 Teams spend hours reading responses manually ⚠️ Real insights get missed So product decisions? Still based on assumptions. That’s where Elvan comes in. Elvan turns customer feedback into a real-time decision engine. ⚡ Here’s how: 🎯 Capture feedback exactly when users experience something 🔗 Bring inputs from product, email, CRM, and more into one place 🤖 Let AI detect sentiment, group patterns, and highlight what matters 🚨 Instantly spot churn risks and feature gaps 💬 Send insights directly to your team in real time No digging through CSVs. No guessing what users mean. Just clear, actionable insights. 💡 The real win? You stop reacting late… and start acting at the right time. 🔗 Product Link: producthunt.com/posts/elvan-3 🌐 Website: elvan.ai 🎁 Early users get direct access to shape the roadmap and share feedback. @elvan_hq #CustomerFeedback #ProductManagement #SaaS #Startups #AI #ProductLedGrowth #CX #UserExperience #Growth #Founders #Tech #BuildInPublic

Launching Elvan on Product Hunt tomorrow. Quick backstory: Qualtrics bought Delighted in 2021. Ignored it for 3 years. Now they're shutting it down June 30 and pushing users to a $420/month enterprise platform. Thousands of SaaS teams, CX leads, and founders now need to find a replacement fast. We built Elvan for exactly that person. NPS, CSAT, CES, eNPS, PMF surveys. AI summaries that tell you what your customers are actually saying. Zendesk integration. Slack alerts. Setup in under 20 minutes. $49/month or free. No enterprise bloat. No onboarding calls. Just signal. Dropping tomorrow on Product Hunt. Would mean a lot if you showed up and upvoted.

@hrprtsingh9 You're right. Listings don't close deals conversations do. We have 500 Delighted users in sequence right now. Ravi (co founder) is personally handling every warm reply. That's the real channel.

@CloudKnight002 Marketplaces are not distribution. Pick up the phone and call 20 buyers this week or the listing will decide for you.

@KaiXCreator Working on Elvan, a feedback platform for SaaS teams. NPS, CSAT, CES surveys with AI that turns responses into plain-English insights. No analyst needed, no enterprise pricing. elvan.ai

@sflorimm Building Elvan, an AI-powered feedback platform for SaaS teams. Think NPS, CSAT, CES with plain-English AI insights so small teams don't need an analyst to make sense of their data. elvan.ai

Looking to connect with people building in: 🍽️ SaaS 🚀 Tech 📲 Automation 🧠 AI tools 📱 Product Development 🔥 Web APP 💻 Devs Drop what you're working on👇

Most “RAG apps” are just thin wrappers around an LLM: Retrieve → Stuff context → Generate → Hope for the best. That’s not AI engineering. That’s prompt plumbing. So I built a corrective RAG system using LangGraph that treats generation as a controlled, stateful process — not a single API call. Here’s what actually happens: • Documents are retrieved from Qdrant • Each document is graded for relevance using an LLM • Low-quality retrieval triggers automatic query rewriting • Generation only runs on filtered, high-signal context • A hallucination check validates grounding • A final answer check ensures the response resolves the user’s question • If validation fails, the system retries intelligently No blind generation. No unvalidated output. No silent failure modes. On top of that, everything is fully observable via LangSmith: node transitions, token usage, retry loops, latency — every step traceable. Production AI is not about better prompts. It’s about orchestration, evaluation, and failure control. That’s AI engineering. #RAG #LangGraph #LangSmith #AIEngineering #LLM

Most “RAG apps” are just thin wrappers around an LLM: Retrieve → Stuff context → Generate → Hope for the best. That’s not AI engineering. That’s prompt plumbing. So I built a corrective RAG system using LangGraph that treats generation as a controlled, stateful process — not a single API call. Here’s what actually happens: • Documents are retrieved from Qdrant • Each document is graded for relevance using an LLM • Low-quality retrieval triggers automatic query rewriting • Generation only runs on filtered, high-signal context • A hallucination check validates grounding • A final answer check ensures the response resolves the user’s question • If validation fails, the system retries intelligently No blind generation. No unvalidated output. No silent failure modes. On top of that, everything is fully observable via LangSmith: node transitions, token usage, retry loops, latency — every step traceable. Production AI is not about better prompts. It’s about orchestration, evaluation, and failure control. That’s AI engineering. #RAG #LangGraph #LangSmith #AIEngineering #LLM

Most “RAG apps” are just thin wrappers around an LLM: Retrieve → Stuff context → Generate → Hope for the best. That’s not AI engineering. That’s prompt plumbing. So I built a corrective RAG system using LangGraph that treats generation as a controlled, stateful process — not a single API call. Here’s what actually happens: • Documents are retrieved from Qdrant • Each document is graded for relevance using an LLM • Low-quality retrieval triggers automatic query rewriting • Generation only runs on filtered, high-signal context • A hallucination check validates grounding • A final answer check ensures the response resolves the user’s question • If validation fails, the system retries intelligently No blind generation. No unvalidated output. No silent failure modes. On top of that, everything is fully observable via LangSmith: node transitions, token usage, retry loops, latency — every step traceable. Production AI is not about better prompts. It’s about orchestration, evaluation, and failure control. That’s AI engineering. #RAG #LangGraph #LangSmith #AIEngineering #LLM

Most AI apps are just chat wrappers. So I built something different — an Agentic AI Control Center using Vercel AI SDK v5. Not just responses. Tool calling. Memory. Structured outputs.

Big lesson: Agentic systems aren’t about better prompts. They’re about orchestration, validation, retrieval design, and architecture. Also — I’m currently building a more advanced RAG application. More on that soon. #AgenticAI #AIEngineering #RAG #Vercel #LLM #BuildInPublic

Fully multi-modal: • Text → Image generation • Audio → Text transcription • Text → Speech synthesis • Context-aware conversational AI Everything unified in one interface.

Most AI apps are just chat wrappers. So I built something different — an Agentic AI Control Center using Vercel AI SDK v5. Not just responses. Tool calling. Memory. Structured outputs.

I built a production-oriented RAG chatbot for querying private documents (policies, internal docs, PDFs). What mattered to me wasn’t “chat with PDFs” — it was retrieval quality and control. Key details behind the scenes: • Admin-only document ingestion using Clerk • Explicit chunking strategy (size + overlap) for semantic coherence • Vector storage with indexed similarity search • Retrieval gated by top-K + similarity threshold to avoid weak context injection • Streaming responses via Vercel AI SDK One small but important choice: I don’t blindly fetch chunks. Retrieval is filtered so only semantically strong context reaches the model — this alone reduces hallucinations more than prompt tweaks. Video + code snippets show: – the API/tool layer – chunking + vector schema – how retrieval feeds generation Repo and live demo are kept private to avoid API misuse and inference cost spikes. Happy to walk through the architecture and trade-offs. #AgenticAI #RAG #AIEngineering #GenAI #BuildInPublic

I’ve been quiet here for ~2 months. Not because I stopped building — but because I was deep into agentic AI. This week, I’m sharing my first Agentic AI project 👇 I built a RAG-based chatbot using Vercel AI SDK where: Admins can upload PDFs (company policies, internal docs, etc.) Users can query those documents conversationally What I focused on (beyond “it works”): • Auth via Clerk (admin-only ingestion) • Thoughtful chunking strategy • Embeddings + semantic search with custom similarity scoring • Clean separation of ingestion, retrieval, and generation logic • Production-ready structure (not a notebook demo) This project wasn’t about stacking tools. It was about understanding why RAG systems behave the way they do. Over the last couple of months, I’ve been learning and experimenting with: RAG, MCP, A2A, Vercel AI SDK, LangChain, LangGraph, LangSmith This is just the baseline. Next up: More agentic systems with reranking, advanced retrieval, and multi-agent workflows using LangChain + LangGraph. If you’re building RAG or agentic AI systems, let’s connect and compare notes. #AgenticAI #RAG #VercelAI #LangChain #LangGraph #AIEngineering #GenAI #BuildInPublic

@3bitslost Appreciate it! Honestly, the biggest leap was moving the whole project to Firebase. Real-time DB, Auth, Hosting, and a proper editor changed everything. Also rewrote the JS file properly instead of the messy 1st-year version.

@Eternalknigh Love this mindset , revisiting old projects is such a good “growth mirror” as a dev. The new version looks much more polished. What was the biggest thing you did differently this time? 👀

In my 1st year, I built a small blog using basic HTML/CSS/JS. Nothing fancy — just my first attempt at a “full-stack” project. I went back to it this week… and decided to rebuild the whole thing. 👇

This rebuild showed me how much I’ve grown as a developer. Sometimes revisiting old projects tells you more than starting new ones. Here’s the updated version: my-mini-app-7f2e2.web.app #javascript #webdev #buildinpublic

I migrated everything to Firebase: • Firestore (real-time DB) • Firebase Auth • Hosting • Analytics • A full blog editor with create/edit/delete Turned a simple project into a proper mini-app.