Writing · index
Notes from the build
Essays, not products. How I approach a problem, what broke, and what I'd do differently.
94 posts · 12 topics
Browse topicsRecent
- Telling users the truth about ingest time
- Flat image in, living character out
- Grounded answers or honest refusals: a rule for retrieval over personal files
- A second legal system: what breaks when a legal pipeline changes country
- What happens when the AI goes down mid-session
- Emotion detection that must never slow the voice
Everything, by topic
Startups, career and moving abroad
17 posts- The AI feature that cut churn, and the one that didn't
- What the buyer actually bought
- Building a product alone: what I automated, what I did by hand
- Leaving a job to build: the first two weeks
- Ideas are cheap. Working software isn't.
- What I look for when hiring an ML engineer
- Why I move between fields on purpose
- Three things big tech does that startups should copy, and three they shouldn't
- What is the Google Project Management Professional Certificate program?
- Attending to Scientist2Entrepreneur (S2E) Program! Finding my way into the world of startups!
- Coping with life after immigration :) ★
- Graph Neural Networks- Architectures review! ★
- An Online Automatic Corona Diagnose System Based on Chest X-ray Images ★
- New Publication: Medical Image Enhancement and Deblurring
- New Publication: An Automatic Driver Assistant based on Intention Detecting Using EEG Signal
- Neural Networks- Review
- الکترومایوگرافی|مقدمه ای بر مهندسی پزشکی فارسی
Shipping models to production
14 posts- What happens when the AI goes down mid-session
- Leakage: the bug that makes every metric lie
- Class imbalance in clinical data: what actually worked
- Most AI never leaves the notebook
- Explaining a model to a doctor
- Monitoring a model that listens to patients
- Cloud Run or EC2: how I chose for a model that listens
- The clinician said no, and was right
- What "in production" means in a hospital
- The boring checklist that saved every project
- Running AI on your own machines
- The baseline before the fancy model: medical image segmentation
- Experiment tracking I actually keep using
- Federated learning: the model was the easy part
Voice AI and companions
14 posts- Emotion detection that must never slow the voice
- Barge-in is the whole product ★
- Why the service speaks English to everyone
- An adversarial persona suite: how I attack my own chatbot
- Fail closed: designing an AI for people who can't afford a bad answer ★
- RAG with a persona: when the knowledge base and the character disagree
- Rewriting a Next.js backend in Python without users noticing
- Streaming speech: where the latency actually goes
- A voice assistant in a weekend: LLM to speech over WebSockets
- Speech to facial animation: a week with Audio2Face
- Real-time transcription on a laptop: what Whisper gets wrong
- Chit Chat Charm ^__^
- TeTrA: Jump to speech controlled VR env
- WordUp! Pro Version Experience!
Retrieval and legal AI
10 posts- Telling users the truth about ingest time
- Grounded answers or honest refusals: a rule for retrieval over personal files
- A second legal system: what breaks when a legal pipeline changes country
- Citations that point at the wrong paragraph
- When BM25 beat the embeddings
- How I evaluate retrieval before anyone asks a question
- Laws have versions, and most RAG pretends they don't ★
- Ingesting a country's law when every province has its own format
- From medical records to legal documents: what changed
- Rules vs transformers for extracting metadata from legal documents
Brain computer interfaces
8 posts- GCN_for_EEG, five years later ★
- BCIGameJam'21! MuSAE Games :) A Winner! ★
- Biological Signal Processing: ML Approach- E01: Breast Cancer Detection ★
- COVID-19 Detection with ML and CXR | تشخیص کرونا توسط یادگیری ماشین و CXR ★
- Open-BCI: First touch
- BCIپردازش تصورحرکتی + راه اندازی آردوینو فارسی
- اضافه کردن EEGLab به متلب و کار با آن فارسی
- معرفی و دانلود دیتابیس ECG و EEG فارسی
Machine learning and AI
8 posts- Recommending books by mood: a small LLM product in a day
- Classical ML or an LLM? The decision before touching code
- Data that doesn't match across sites
- Audio preprocessing that matters more than the model
- Respiratory sound classification, end to end ★
- What an earbud's motion sensor can tell you
- Doing ML in a field that runs on R and SPSS
- Using HuggingFace without sharing your code!