The production-grade, cohort-based engineering fellowship for developers stepping into LLMs, vector systems, MCP agents, and MLOps. Learn live, ship hands-on labs, and build a defense-ready capstone.
Every weekend deepens your engineering depth. Live sessions on Google Meet, code-alongs in Colab, and graded labs with instructor feedback.
Vector calculus, matrix multiplication internals, contiguous memory layouts, NumPy internals, and building a Micrograd-style autograd engine from scratch.
LightGBM and XGBoost feature engineering, population stability index (PSI) drift monitoring, and handling covariate shift at scale.
Multi-head attention and rotary positional embeddings from scratch. Distributed PyTorch DDP pipelines.
Hybrid search with BM25 + dense vectors, Reciprocal Rank Fusion, re-ranking, LangGraph multi-agent state machines, and Model Context Protocol (MCP) servers.
LoRA, QLoRA 4-bit NF4 mechanics, GGUF/llama.cpp deployment, continuous batching, and vLLM PagedAttention KV-cache optimisation.
Dockerised Triton serving, Prometheus SLA telemetry, rate limiters, token counting proxies, and a live technical defense.
Most ML courses teach three lines of SDK boilerplate. We teach you to engineer retrieval pipelines, evaluate models honestly, and ship AI services that survive production.
$ pytest tests/test_rag_pipeline.py -v
PASSED tests/test_rag_pipeline.py::test_hybrid_bm25_dense_rrf [fusion order]
PASSED tests/test_rag_pipeline.py::test_retrieval_latency_budget [p95 within budget]
PASSED tests/test_rag_pipeline.py::test_ndcg_at_10_evaluation [above threshold]
PASSED tests/test_rag_pipeline.py::test_mcp_agent_tool_dispatch [tool schema]
==== 4 passed ====Late in the cohort, fellows take part in a 48-hour build sprint, then defend a production-grade capstone in a live technical review.
Teams ship LangGraph multi-agent systems with self-correction, sandboxed tools and telemetry — judged live on engineering quality.
Bi-encoder embeddings, BM25 keyword matching and cross-encoder re-ranking for fast candidate scoring.
A multi-agent pipeline that parses long financial filings, extracts tables into Postgres and cross-checks claims.
Criteria extraction and matching against de-identified records with a privacy-preserving local model.
Every live session, code review and office hour is run by practising AI engineering leaders — not recorded content.
Leads AI engineering teams building production ML and GenAI systems. Conducts all 24 live deep-dive coding sessions, line-by-line code reviews, and office hours.
No hidden fees and no income-share traps. Simple upfront pricing with a GST invoice for employer reimbursement.
Not ready for the 12-week fellowship? Test-drive the teaching style. Build a live Model Context Protocol (MCP) agent on Google Meet this weekend.
Schedule commitments, compute, refunds and the admissions process.
Sessions run every Saturday and Sunday from 10:00 AM to 12:30 PM IST (2.5 hours per session) on Google Meet. Expect another 4 to 6 hours during the week for labs and office hours.
Every session is recorded and published to your learner portal alongside notebooks, code and reading links. You keep lifetime access.
You should be comfortable with intermediate Python (functions, classes, comprehensions) and basic linear algebra. Prior ML experience is not required — Module 01 builds the fundamentals.
No. All labs are designed to run on Google Colab, RunPod, or serverless endpoints. Setup instructions are provided on Day 1.
Full refund if requested before Session 02 begins. If work interferes, you can defer your seat to the next batch at no extra charge.