Anjana - AI & Tech

Asked in a Swiggy AI/ML Engineer interview 👇 You have a vector database containing 10 million documents. How would you retrieve the 5 most relevant documents without comparing the query against all 10 million documents? The interviewer is checking one thing: do you know exact search does not scale, and can you name what replaces it? The answer is approximate nearest neighbour search — you trade a sliver of recall for a thousandfold speedup. 1. Say the cost of brute force out loud - 10M vectors × 768 dimensions is a full scan on every single query - Exactly correct, and unusable at any real QPS 2. HNSW — the default - A layered graph: sparse top layers for long jumps, a dense bottom layer for precision - Enter at the top, greedily walk towards the query, drop a layer, repeat - Visits a few thousand nodes instead of 10 million - efSearch is your recall-versus-latency dial at query time 3. IVF — partition and prune - Cluster the vectors into centroids; search only the nprobe nearest cells - Faster to build and lighter in memory than HNSW, at slightly lower recall 4. Compress so it fits in RAM - Product quantisation shrinks a vector 10-30x; disk lookups are what actually kill latency - IVF-PQ for huge corpora, then rescore the shortlist with full-precision vectors 5. Filter inside the search, not after it - Tenant, date and language as pre-filters on the index - Post-filtering silently hands back fewer than 5 results 6. Over-fetch, then rerank - Pull top-50 with ANN, rerank with a cross-encoder, return 5 - Recovers most of the accuracy the approximation gave away The principle: nobody searches 10 million vectors. You search a structure built over them, and you choose how much recall to trade for latency. 💬 HNSW or IVF-PQ for your workload — and what made you pick it?

Asked in a Spotify AI/ML Engineer interview 👇 A user tells Spotify, “Play something relaxing for my evening commute.” How would you design an AI system that understands the request and creates a personalized playlist based on mood, history, and context? There are three requests hidden in that sentence: a mood, a duration and a setting. Translate it into constraints, then let retrieval and sequencing do the work. 1. Parse the request into constraints, not keywords - "Relaxing" maps to audio features: low energy, moderate tempo, acoustic, low speechiness - "Commute" implies a length — build to roughly 35 minutes, not an endless queue - "Evening" narrows the listening history you should trust 2. Let the LLM interpret, never invent - The model turns language into filters and weights; the catalogue supplies the tracks - A hallucinated song title is unplayable, and users notice immediately 3. Ground everything in this listener - Blend familiar tracks from their own library with new ones, and tune that ratio per user - Their "relaxing" is learned from what they actually played on past evenings, not from a genre tag 4. Sequence the playlist, do not just select it - Order is the product: ease in, hold the mood, land softly before the commute ends - No abrupt tempo or key jumps, and a cap on tracks per artist 5. Fill the gaps with sensible defaults - Nothing stated about genre? Use what they played at this hour on this weekday - Cold start falls back to editorial seeds for the mood, then adapts within minutes 6. Learn from the skips - A skip inside 30 seconds is a strong negative; completions and saves are positives - Re-rank the unplayed tail mid-session rather than waiting for the next request

15 Machine Learning Projects (Step by Step Guide) along with Source Code👇 https://www.instagram.com/reel/DegGsclyJJU/?xtok=MWFvazJseHk1ZXRzMA==

Asked in a Meta AI Engineer interview 👇 How would you design a GenAI system capable of generating 50 million personalized ad copies every day? How would you ensure each copy feels handcrafted and does not appear AI-generated? Do the arithmetic first. 50 million a day is about 580 copies every second, forever. Nobody calls a frontier model 580 times a second and survives the bill. 1. Generate templates, not copies - The model produces a few thousand reusable skeletons per campaign, offline - Serving fills the slots from user features: city, last purchase, price band - Generation becomes a batch job; the request path is a lookup 2. Segment instead of personalising per user - Cluster users into a few thousand behavioural cohorts - One creative set per cohort × product × placement - 50 million impressions, far fewer distinct creatives 3. Kill the AI tells - Ban the tics: "unlock", "elevate", "in today's fast-paced world" - Vary sentence length deliberately — uniform rhythm is the giveaway - Sample with temperature and a per-copy style seed, never greedy decoding 4. Get voice from retrieval, not adjectives - Retrieve the brand's best human-written ads as few-shot examples - "Write in a friendly tone" produces the same beige copy for every brand 5. Filter before anything ships - Near-duplicate detection over the whole day's output with embeddings and MinHash - Claim and compliance checks: prices, guarantees, regulated categories - A classifier trained on your own human copy scores "does this sound like us" 6. Let the market judge - Multi-armed bandits on live CTR; kill losers in hours, not weeks - Winners re-enter the few-shot pool, so the system compounds The principle: at 50 million a day you are not generating copy, you are generating a distribution. Handcrafted means varied, on-voice and evaluated — not expensive. 💬 Templates plus slots, or per-user generation — which way would you go?

Asked in a Netflix AI Engineer interview 👇 A Netflix user watches a movie and says, “Show me something like this, but more exciting.” How would you design a GenAI recommendation system that understands the user’s intent and generates personalized recommendations? "More exciting" is a direction, not a query. The request is an anchor plus a vector, and a plain similarity search answers the wrong question. 1. Parse the instruction into anchor, axis and constraints - Anchor: the title they just watched - Axis and magnitude: pace and intensity, moved up a notch - What must not change: genre, language, tone, maturity rating 2. Label the catalogue on the axes people actually use - Pace, intensity, humour, complexity, ending type, violence - Derive them from metadata, subtitles and trailers; LLM-tag what is missing - Human-verify the axes that drive the most requests — this is the real work 3. Move the vector, then retrieve - Take the anchor embedding, add a delta along the excitement axis, run ANN over the catalogue - Then apply availability, language and maturity filters 4. Personalise the result, do not replace it - Blend with the member's taste model so two people asking the same thing get different rows - The instruction constrains, the taste model orders 5. Generate the explanation, never the item - The LLM writes "faster paced, same heist premise"; the catalogue decides what exists - Ungrounded generation recommends titles you do not have the rights to 6. Keep the session open - "Even more" should travel further along the same axis, not reset - Log every direction request — they are free, precise preference labels

Become a Job Ready Data Engineer(In Just 7 Weeks) Week 1: Learn SQL Beginner to Advanced in Under 4 Hours https://www.youtube.com/watch?v=OT1RErkfLNQ Week 2: Learn Python - Full Course for Beginners [Tutorial] https://www.youtube.com/watch?v=rfscVS0vtbw Python for Data Science - Course for Beginners (Learn Python, Pandas, NumPy, Matplotlib) https://www.youtube.com/watch?v=LHBE6Q9XlzI Week 3: Data Modeling Masterclass For Data Engineers https://www.youtube.com/watch?v=K7C1sWKQU-o Apache Airflow One Shot- Building End To End ETL Pipeline Using AirFlow And Astro https://www.youtube.com/watch?v=Y_vQyMljDsE Week 4: Databricks Tutorial | Databricks Free Edition Tutorial with End-to-End Data + AI Project https://www.youtube.com/watch?v=761SQ9Hxbic PySpark Tutorial | Full Course (From Zero to Pro!) https://www.youtube.com/watch?v=94w6hPk7nkM Week 5: [Data Build Tool] DBT - The Ultimate Guide | With CI/CD https://www.youtube.com/watch?v=B8uwFmVt4sU Week 6: AWS Masterclass for Data Engineers with Project https://www.youtube.com/watch?v=yvAWbbQa8eE Azure Data Engineer Full Course | Azure Data Factory Tutorial For Beginners https://www.youtube.com/watch?v=YkK0mdcJfHg Week 7: End to End Data Engineering Project using Databricks Free Edition | FMCG Domain https://www.youtube.com/watch?v=U6ZUKWdfSLY Zomato AI Data Analytics | End-To-End AI Data Engineering Project https://www.youtube.com/watch?v=kYwaNMQ3XT8

Asked in an Apple AI/ML Engineer interview 👇 How would you design an AI assistant that can read a user’s messages, calendar, and emails to prepare them for a meeting? How would you make the assistant automatically become silent when the phone is handed to another person? These are two systems, not one. The briefing is a retrieval problem. The silence is an identity problem, and that is the half most candidates skip. 1. Keep the corpus on the device - Messages, mail and calendar are the most sensitive data a person owns - On-device index and a small model by default; cloud only for what cannot run locally - If a request must leave the device, strip names and send the minimum 2. Build the briefing from the calendar, not a search box - Trigger off the event: attendees, time, title - Per attendee, pull the last threads, open commitments and shared documents - Emit a fixed shape: who they are, what changed since last time, open asks, your talking points 3. Rank by obligation, then recency - Unanswered questions addressed to the user outrank everything - Decay old threads — a six-month-old email is background, not news 4. Treat the handoff as continuous authentication - Face and voice presence checked continuously, not once at unlock - Grip, gait and motion changes from the accelerometer flag a change of hands - Any drop in identity confidence locks personal context immediately 5. Fail closed, and fail quietly - Uncertain identity means generic answers or nothing at all - Never announce what it is hiding: "you have a private message" leaks the message - Recovery is explicit — the owner re-authenticates and context returns 6. Make the boundary inspectable - One clear indicator for "personal context on" - A log of what was read and when, that the user can review and revoke The principle: personalisation is a permission, not a feature. The assistant must know who it is talking to before it decides what to say. 💬 Which signal would you trust most to detect a handoff?