Specialization + Applications

All 5 Phases

Week-by-Week Schedule

Topics Covered

Every subtopic below is a separate daily task in the roadmap, with hand-picked resources (YouTube videos, docs, papers) for each.

• vLLM: continuous batching, PagedAttention
• Ollama for local LLM serving
• Quantization: GPTQ, AWQ, GGUF, INT4
• AWS basics: S3, EC2, IAM
• AWS Bedrock for hosted LLM access
• Stand up vLLM locally with a small model, benchmark tokens/sec
• Deploy a model to AWS (EC2 or Bedrock) — get hands-on cloud experience
• LLM-as-a-judge: biases (verbosity, position, self-preference)
• Golden datasets + regression suites
• DeepEval + Promptfoo for systematic testing
• Prompt injection attacks + defenses (OWASP LLM Top 10)
• Guardrails: input/output filtering, allowlists
• Add full eval suite + CI gates to your RAG project
• Add prompt injection guardrails + red-team your own agent
• Find a "good first issue" on a target repo
• Clone, set up dev environment, understand codebase
• Work on the issue
• Work on the issue
• Open the PR
• Start CAPSTONE: Multi-agent system with evals
• Continue capstone
• Wed — Build core agent + tool integrations
• Pydantic validation on all tool inputs
• Build golden eval set
• Set up CI/CD with eval gates
• Deploy + Langfuse tracing — Portfolio Capstone Project #5
• Identify 30 target companies (5 stretch MAANG, 15 strong AI startups, 10 realistic)
• Apply to 10 — tailor resume per role
• LinkedIn outreach: 20 AI engineers at target companies
• Ask 3 connections for referrals
• Apply to 10 more
• Write 3 detailed blog posts on your projects (Medium/personal site)
• Cross-post projects on Twitter/X with #BuildInPublic
• DSA: 1 medium problem to stay sharp
• Review ML breadth flashcards (Anki)
• Prep for upcoming interviews (research company + recent papers from their team)
• DSA: 1 medium problem + ML flashcards
• DSA + flashcards + interview prep continued
• Mock interview + project deep-dive practice
• Mock interview + project deep-dive practice
• Track A — Applied LLM Engineering: agents, MCP, structured generation, observability
• Track B — ML Platform / MLOps: Kubernetes, Ray, feature stores (Feast)
• Track C — Multimodal AI: CV + LLMs, VLMs like LLaVA, Qwen-VL
• Track D — Foundation Models / Research (needs MS+); read papers, reproduce SOTA
• Pick ONE track and outline mini-project scope
• Deep-dive study on chosen track
• Start mini-project in chosen specialization
• GRE prep starts (target 325+, Quant 168+) — Magoosh or Manhattan Prep
• GRE prep continues
• GRE prep + identify 10 target universities (Tier 1/2/3)
• GRE prep + identify 5 faculty per university whose research aligns
• Draft SOP v0 — concrete research alignment, not generic AI enthusiasm
• Buffer: catch up on incomplete weeks, polish capstone
• Plan next 6 months: MS timeline + specialization deep-dive