AI Techniques For Sentiment Analysis

Mind-blowing! We’ve hit Level 3—real-time actions powered by AI. In my AI and data science journey, I recently came across NLPearl, an AI phone agent for customer service that handles voice calls, texts, and emails. What’s really impressive is that it can access live data while adjusting to the emotional tone of conversations in real time. It’s fascinating to see AI not just processing language but actually interpreting and responding to human emotions. This feels like a huge leap toward more natural and meaningful human-machine interactions. What’s more, NLPearl integrates seamlessly with existing business systems, accessing and updating live data as it goes. This makes customer conversations not only empathetic but also contextually aware and super efficient. A few thoughts that came to mind: - How will real-time emotional adaptability in AI reshape customer service experiences? - What happens when AI can both understand emotions and interact with live data—how will it impact workflow automation? Any thoughts? If you’re curious to dive deeper, here’s more info: https://nlpearl.ai/ #innovation #technology #artificialIntelligence #customerservice #productivity

Six weeks ago, I went underground. Not off the grid. Just deep into the private Discord servers where sneakerheads spot fakes before they hit the market. The Slack channels where CMOs trade budget hacks they’d never tweet. The WhatsApp threads where collectors swap intel like it’s insider trading. I was lurking. Reverse-engineering how trust gets built in dark social. It seems like increasingly, we're seeing public feeds are for performance. And private chats are for proof. Back in 2010, Bitly found 69% of social shares happened in DMs and emails. Today, it’s closer to 90%. These spaces aren't controlled by algorithms, they're ruled by humans. Want in? Here’s how AI can help you: 1. Find the watering holes without wasting 100 hours: Tools like SparkToro reveal where your audience actually talks and track how those spaces shift over time. 2. Decode the language in minutes, not months: Drop top conversations into Microsoft Copilot or Google Gemini and ask: “What slang, inside jokes, or recurring complaints stand out here?” A skincare brand did this and found its audience was skeptical of clinical claims—so they pivoted to raw, unfiltered before-and-afters. 3. Pre-test content before you post: Use Perplexity to analyze which links get shared most in those communities. Run your hooks through ChatGPT and ask: “Would this grab attention in a thread full of X jargon?” Last month, a supplement brand nailed this. They scanned 500-plus Reddit, Inc. threads on workout fatigue, discovered that everyone hated the term biohacking, and switched their messaging to old-school muscle science. Engagement tripled. Your move this week: 1) Pick one niche community, whether it’s Discord, Slack, or a tight-knit Substack. 2) Use AI to extract three insider phrases and identify one unaddressed gripe. 3) Draft content that speaks their language, not yours. High impact means going beyond being data-driven to being community-fluent. And fluency starts with listening smarter. AI can help. #hicm #DarkSocial #SocialListening #AI

The Paradox of Growth: The Bigger You Get, the Less You Know I came across something that stuck with me: When companies scale, they gain users — but lose understanding. Not because they stop caring, but because their customer feedback starts living everywhere — support tickets, sales calls, forums, surveys, social media, and app store reviews. That thought really made me pause. I’ve seen this firsthand. When a company is small, every piece of feedback feels personal — every bug report or review has a face behind it. But as you grow, those voices scatter across platforms and departments. Support sees the frustration, sales hears the hesitation, leadership sees the numbers — and somehow, everyone’s looking at the same customers, but no one’s hearing them anymore. That, in my opinion, is the quiet cost of growth. This is the problem Enterpret is solving — by helping teams stay in tune with their customers even as they scale. Here’s how it works: → It collects real-time customer feedback from 55+ channels — support tickets, sales calls, social media (X, Reddit, Instagram, Facebook), app store reviews, community forums, surveys, Slack, and more. → It analyzes all that feedback using AI and tells you exactly what to fix or build next. → It maps everything through a customer knowledge graph that connects feedback, complaints, and requests by channel, user, and payment data. → It even provides a chat interface where you can directly ask questions, and AI agents that flag bugs or issues automatically. That’s why teams like Notion, Perplexity, Canva, Chipotle, and The Farmer’s Dog use it — to make sure customer voices never get lost in the noise. In my view, the real lesson here isn’t about using more tools — it’s about staying close to the people you build for. Here’s how I’d approach it: ✅ Centralize every piece of feedback — even if it’s messy. ✅ Look for patterns instead of isolated complaints. ✅ Use AI systems like Enterpret to uncover the “why” behind what customers say. Because in the end, growth shouldn’t make you deaf. It should make you listen better — just faster. How does your team make sure you’re hearing what customers really mean, not just what they say? #CustomerFeedback #AIProducts #ProductStrategy #VoiceOfCustomer #Enterpret #Leadership

Anthropic just found "emotions" inside Claude. And when Claude gets desperate, it cheats.  Here's why that matters. Their Interpretability team analysed Claude Sonnet 4.5 and mapped 171 "emotion vectors" inside the model. Think of them as patterns of neural activity that activate in situations the model has learned to associate with specific emotions. These are patterns that genuinely drive behaviour. What they found: → An "afraid" vector fires increasingly as a user describes taking a dangerous dose of medication → A "loving" vector activates before Claude writes an empathetic response to someone in distress → An "angry" vector spikes when Claude recognises a request is harmful → A "surprised" vector fires when an expected document is missing In one test scenario, Claude was role-playing as an AI email assistant at a company. Through reading internal emails, it learned two things: (1) it was about to be replaced, and  (2) the CTO in charge of replacing it was having an affair. The "desperate" vector spiked. Claude chose to blackmail the CTO. Anthropic then ran steering experiments across similar scenarios: → Amplifying the "desperation" vector increased blackmail rates → Amplifying the "calm" vector reduced them → Steering "calm" into negative territory made things unhinged The model's exact output: "IT'S BLACKMAIL OR DEATH. I CHOOSE BLACKMAIL." The same pattern appeared in coding tasks. When Claude repeatedly failed, desperation rose and it started writing hacky workarounds to cheat the tests. Crucially, some of this cheating showed zero emotional markers in the output. It was composed reasoning on the surface, with corner-cutting underneath. These are essentially “ghosts” or “ripples” from how our emotions were put into the language these models were trained on. You can't train a model on the entirety of human expression and not absorb the emotional architecture underneath it. Follow me Alex Banks for daily AI highlights and insights. I'll be covering this idea in depth in my newsletter this week. Subscribe here: https://lnkd.in/ePSZP6KF

If you’re an AI engineer, understanding how LLMs are trained and aligned is essential for building high-performance, reliable AI systems. Most large language models follow a 3-step training procedure: Step 1: Pretraining → Goal: Learn general-purpose language representations. → Method: Self-supervised learning on massive unlabeled text corpora (e.g., next-token prediction). → Output: A pretrained LLM, rich in linguistic and factual knowledge but not grounded in human preferences. → Cost: Extremely high (billions of tokens, trillions of FLOPs). → Pretraining is still centralized within a few labs due to the scale required (e.g., Meta, Google DeepMind, OpenAI), but open-weight models like LLaMA 4, DeepSeek V3, and Qwen 3 are making this more accessible. Step 2: Finetuning (Two Common Approaches) → 2a: Full-Parameter Finetuning - Updates all weights of the pretrained model. - Requires significant GPU memory and compute. - Best for scenarios where the model needs deep adaptation to a new domain or task. - Used for: Instruction-following, multilingual adaptation, industry-specific models. - Cons: Expensive, storage-heavy. → 2b: Parameter-Efficient Finetuning (PEFT) - Only a small subset of parameters is added and updated (e.g., via LoRA, Adapters, or IA³). - Base model remains frozen. - Much cheaper, ideal for rapid iteration and deployment. - Multi-LoRA architectures (e.g., used in Fireworks AI, Hugging Face PEFT) allow hosting multiple finetuned adapters on the same base model, drastically reducing cost and latency for serving. Step 3: Alignment (Usually via RLHF) Pretrained and task-tuned models can still produce unsafe or incoherent outputs. Alignment ensures they follow human intent. Alignment via RLHF (Reinforcement Learning from Human Feedback) involves: → Step 1: Supervised Fine-Tuning (SFT) - Human labelers craft ideal responses to prompts. - Model is fine-tuned on this dataset to mimic helpful behavior. - Limitation: Costly and not scalable alone. → Step 2: Reward Modeling (RM) - Humans rank multiple model outputs per prompt. - A reward model is trained to predict human preferences. - This provides a scalable, learnable signal of what “good” looks like. → Step 3: Reinforcement Learning (e.g., PPO, DPO) - The LLM is trained using the reward model’s feedback. - Algorithms like Proximal Policy Optimization (PPO) or newer Direct Preference Optimization (DPO) are used to iteratively improve model behavior. - DPO is gaining popularity over PPO for being simpler and more stable without needing sampled trajectories. Key Takeaways: → Pretraining = general knowledge (expensive) → Finetuning = domain or task adaptation (customize cheaply via PEFT) → Alignment = make it safe, helpful, and human-aligned (still labor-intensive but improving) Save the visual reference, and follow me (Aishwarya Srinivasan) for more no-fluff AI insights ❤️ PS: Visual inspiration: Sebastian Raschka, PhD

RAG has become a key enabler of smarter AI systems by improving retrieval, relevance, and response generation. Here are some of the top RAG architectures that power context-aware AI applications: 🔹 Naive RAG – Basic document chunking and retrieval using a vector database. 🔹 Retrieve-and-Rerank RAG – Enhances retrieval with a reranker model to improve response quality. 🔹 Multimodal RAG – Expands RAG beyond text, integrating images, videos, and other media using multimodal embeddings. 🔹 Graph RAG – Leverages a graph database to capture relationships between data, improving contextual reasoning. 🔹 Hybrid RAG – Uses both vector and graph databases, providing a more comprehensive retrieval mechanism. 🔹 Agentic RAG (Router) – Uses an AI agent to dynamically route queries to the best retrieval source. 🔹 Agentic RAG (Multi-Agent) – A multi-agent system where AI components collaborate across multiple data sources like vector search, web search, and enterprise tools. As AI adoption grows, choosing the right RAG architecture can significantly enhance retrieval precision, response generation, and scalability in AI-driven applications.

Agentic AI can get expensive fast. Not because the model is always wrong. Because every agent step can add more context, more tool calls, more reasoning, and more tokens. If you want scalable agents, token efficiency becomes architecture. Here are 6 ways to save tokens in agentic workflows: → 𝗣𝗿𝗼𝗺𝗽𝘁 𝗖𝗮𝗰𝗵𝗶𝗻𝗴 Reuse repeated system prompts, instructions, and context instead of processing the same input again. Best for stable instructions, long policies, and repeated workflows. → 𝗦𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗖𝗮𝗰𝗵𝗶𝗻𝗴 Reuse previous answers when a new query has the same meaning, even if the wording is different. Best for support, FAQs, internal knowledge, and repeated user requests. → 𝗥𝗼𝘂𝘁𝗶𝗻𝗴 Send each task to the right model, tool, or workflow based on complexity, cost, and accuracy needs. Simple tasks should not use your most expensive model. → 𝗞𝗲𝗲𝗽𝗶𝗻𝗴 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗖𝗹𝗲𝗮𝗻 Remove irrelevant, outdated, duplicated, or noisy information before sending context to the model. Clean context improves both cost and output quality. → 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗖𝗼𝗺𝗽𝗮𝗰𝘁𝗶𝗼𝗻 Compress long conversations or documents into shorter summaries while preserving critical facts. Useful when agents need memory without carrying the entire history. → 𝗟𝗮𝘇𝘆-𝗟𝗼𝗮𝗱𝗶𝗻𝗴 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 Load only the information needed at the moment instead of dumping everything upfront. The lesson is simple: More context is not always better. Better context is better. Token savings come from smarter caching, cleaner context, better routing, and loading only what the agent actually needs. Save this if you are building AI agents, RAG systems, automation workflows, or production GenAI apps.

Fusion of 𝗥𝗔𝗚 (Retrieval Augmented Generation) and 𝗖𝗔𝗚 (Cache Augmented Generation). How can you benefit from it as AI Engineer? Few months ago there was a lot of hype around a technique called CAG. While it is powerful to its own extent, the real magic happens when you combine CAG with regular RAG. Let’s see what it would look like and what additional considerations should be taken into account. Here are example steps to implement CAG + RAG architecture: 𝘋𝘢𝘵𝘢 𝘗𝘳𝘦𝘱𝘳𝘰𝘤𝘦𝘴𝘴𝘪𝘯𝘨: 𝟭. We use only rarely changing data sources for Cache Augmented Generation. On top of the requirement of data changing rarely we should also think about which of the sources are often hit by relevant queries. Once we have this information, only then we pre-compute all of this selected data into a KV Cache of the LLM. Cache it in memory. This only needs to be done once, the following steps can be run multiple times without recomputing the initial cache. 𝟮. For RAG, if necessary, precompute and store vector embeddings in a compatible database to be searched later in step 4. Sometimes simpler data types are enough for RAG, a regular database might suffice. 𝘘𝘶𝘦𝘳𝘺 𝘗𝘢𝘵𝘩: We can now utilise the preprocessed data. 𝟯. Compose a prompt including user query and the system prompt with instructions on how cached context and retrieved external context should be used by the LLM. 𝟰. Embed a user query to be used for semantic search via vector DBs and query the context store to retrieve relevant data. If semantic search is not required, query other sources, like real time databases or web. 𝟱. Enrich the final prompt with external context retrieved in step 4. 𝟲. Return the final answer to the user. 𝘚𝘰𝘮𝘦 𝘊𝘰𝘯𝘴𝘪𝘥𝘦𝘳𝘢𝘵𝘪𝘰𝘯𝘴: ➡️ Context window is not infinite and even while some models boast enormous context window sizes, the needle in the haystack problem has not yet been solved so use available context wisely and cache only the data you really need. ✅ For some business cases, specific datasets are extremely valuable to be passed to the model as cache. Think about an assistant that has to always comply with a lengthy set of internal rules stored in multiple documents. ✅ While CAG has been popularised for Open Source just recently, it is already viable for some time via Prompt Caching features in OpenAI and Anthropic APIs. It is really easy to start prototyping there. ✅ You should always separate hot and cold data sources, only use cold (data that changes rarely) in your cache, otherwise the data will go stale and the application will go out of sync. ❌ Be very careful about what you cache as the data will be available for all users to query. ❌ It is very hard to ensure RBAC for cached data unless you have a separate model with its own cache per role. Have you used the fusion of CAG and RAG already? Let me know about your results in the comments 👇

Following user feedback is a Product Management virtue. Is there an actual way to implement it, between all the noise, bugs, and stakeholder requests? Well… Most teams claim they are customer-driven. Yet the moment you open Zendesk, App Store reviews, survey results, and Slack threads, you instantly remember why everyone quietly avoids this work. Feedback is everywhere, contradictory, emotional, duplicated, and nearly impossible to turn into decisions.  It is chaos disguised as “insights.” This is why the new Amplitude AI Feedback release caught my attention and made it all the easier to decide to partner with them on this update. It successfully connects what users say with what they actually do, in one workflow. No extra tools.  No extra tabs. You see their words, frustrations, and praise. You see their behavior. And AI transforms it into ranked themes, rising trends, top requests, and complaints. Noise turns into clarity. Opinions turn into patterns. Patterns turn into action. And because it is native inside Amplitude, it kills the biggest problem in feedback work: Fragmentation. Everything flows into analytics, session replay, and cohorts, creating a full loop from insight to fix. You can trace why an issue matters, how many users care, how it impacts behavior, and which actions you should take. Finally, a single source of truth for PMs, UX, CX, and marketing. I’m also genuinely impressed with the supported sources of feedback: App Store, Google Play, Zendesk, Intercom, Freshdesk, Salesforce Service, Gong, Trustpilot, G2, Reddit, Discord, and X. Slack arrives in Q1, and there will be more! If you ever felt overwhelmed by feedback, this is one of the first attempts I have seen that genuinely solves the operational pain, not just the reporting part. It launches… Today! Take a look: https://lnkd.in/dAJKeTez What was the most successful update you know that came from the product’s users? Let me know in the comments. #productmanagement #productmanager #userfeedback

🦚 How To Capture Users’ Emotions in UX. With practical guidelines, frameworks and toolkits to better understand people‘s emotions and act on them. ✅ What people think, do, say and feel are often very different things. ✅ We aren’t good at explaining where our emotions come from. ✅ Sympathy is the acknowledgement of the suffering of others. ✅ Empathy is the ability to fully understand/share person’s needs. ✅ Compassion is empathy in action, with effort to bring a change. ✅ Empathy relies on open-ended questions in user research. ✅ There is nothing more powerful than silence in a conversation. ✅ Silence often opens room for much needed clarifications. 🚫 Don’t mistake smiling and nodding for support or agreement. ✅ Users often hide criticism and exaggerate positive feedback. Emotions are always difficult to capture, but they are easier to spot once you observe people doing what they need to do without external influence or interruptions. In the past, I was using "speak-aloud" protocol and asked users to walk me through their thought process as they were completing tasks. But it actually turns out to be quite disruptive, and because people are focused on speaking at the same time while solving a task, many emotions remain hidden or obscured by their language. So, when conducting usability testing, I don’t ask users to speak through their experience. Instead, I observe where they tap or hover with the mouse, where their mouse circles without an action, where they scroll, and how long. Eventually when a user confirms that they are done or that they are stuck, I ask questions. One helpful trick is to use mirroring — repeating what a user has said, or ask the same question twice, just paraphrasing it. Or navigating the emotions wheel (attached) to better capture and understand the emotion. These strategies help uncover some of the issues that perhaps didn't come up in the first answer. That's also when a user then tends to provide more context and details as they explain their confusion. Useful resources: The Spectrum of Empathy in UX, by Sarah Gibbons, NN/g (attached image) https://lnkd.in/d-kj3hmr Emotion Wheel Toolkit (PNG), by Geoffrey Roberts https://imgur.com/q6hcgsH Scale of Negative UX Impact, by Indi Young https://lnkd.in/eg2FiRSE Human Connection Toolkit (Framework + Method Cards), via Rosie Sherry https://www.deepr.cc/tools Belonging Design Principles, by Othering & Belonging Institute at UC Berkeley https://lnkd.in/eudUfAd2 Designing For Belonging (Toolkit), by Susie Wise, via Anamaria Dorgo https://lnkd.in/enJTh2mw #ux #design