I’m doing some #freelance AI/ML work this month -small stuff, quick builds, whatever you need done fast. What I’ve worked on lately: • an #AI trading agent (Dazai) • a Market Regime Classifier • a LegalAI chatbot (RAG + FastAPI + Chroma) • a bunch of #ML models + full-stack

Link :

Played turf football with total strangers last night—barely knew anyone. Didn’t even share a language, yet still scored once and had fun. Making connections never needed words, Ig... #LifeMoments

Tried deploying my Lexx project again and ran into Render storage limits. Switching to ngrok for now. Building stuff is mostly just dealing with random constraints n issues.... #DevLife #AIProjects

Didn’t expect this. One of my datasets on #HuggingFace crossed 140+ downloads. Not viral, not huge - but real people found it useful, still means a lot to me... Proof that putting work out there actually compounds. #MachineLearning #Datasets #LearningByDoing

Watched #YC’s discussion on vertical AI agents because of my own Lexx #project. Must watch if you’re building domain-specific #AI -reinforces why agents only work when the problem space is narrow and deeply understood. Helped me rethink a few design choices. #AIAgents

Watching a walkthrough on building a RAG app with Amazon Bedrock + Knowledge Bases. Good reminder that RAG isn’t just about embeddings - chunking, retrieval logic, and infra decisions matter way more than people admit. Learning by watching others wire it end-to-end helps a lot.

Watched the AWS DeepRacer demo while doing it myself today, trained the agent using PPO and SAC - PPO for discrete actions, SAC for continuous control. cool real-world example of how RL algorithms actually differ beyond theory.

Spent multiple days deep in LegalAI (Lexi). Rebuilt parts in Next.js that took forever in Colab — crazy how much cleaner things move locally. Auth + MongoDB are in, chats are saving, but RAG + history still collide. Messy logs, broken flows… classic backend pain. Still

I low-key envy people whose mindset doesn’t shake when they see peers moving ahead. My brain keeps reminding me how much I don’t know… and at the same time, I know I’m ahead in ways that aren’t obvious yet. Weird place to be...🧐

Tried scaling my RAG setup with multiple files today. Reality check: Colab limits slow everything down. You learn infra pain the moment things actually start working.

Reading this paper on how NLP is actually used in the legal system. Good reality check - beyond chatbots, it’s about retrieval, summarisation, judgment prediction, and reducing manual legal grunt work. Slowly connecting the dots for my own #Legal AI ideas.

Saw an Indian bail dataset on Hugging Face - built by a BTech student, basically my age. 1k+ downloads already. while i just discovered Hugging Face and I’m still figuring things out. But that’s fine. I’ll get there too. Not slowing down. Fighting!! Fighting!!

Uploaded one of my datasets to @huggingface . Multi-timeframe BTC market regimes - parquet, structured, ready for experiments. Not even sure I did everything perfectly, but you’ve got to start somewhere. Feels good to finally put work out there.😊

Tried building a medical-specific #LLM as a mini project. I thought it’s mostly fine-tuning… turns out the moment you move into a real domain, everything gets a bit strict- data quality, safety, hallucinations, evaluation. General models feel easy in comparison. Hard project,

Presented the Minor project and one thing became clear.. Evaluations often reward documentation and presentation over actual working depth. Good slides hide weak even stupid project, while solid ideas gets ignored if it’s not packaged well.

Use #Reddit if want quick engagement. Reddit is actually the fastest place to get eyes on your stuff, even if you have no followers anywhere... Great for: – resume reviews – side project feedback – basic life/career advice You just need the right subreddit ... the crowd does