Deep tech,
written and visualized.

LLM architectures, inference servers, sampling, prompt caching, observability.

Attention variants, RoPE, MoE, RMSNorm, MTP, sparse attention, FlashAttention.

RAG, hallucination mitigation, embeddings, evals, prompt engineering, agentic safety.

Tool use, agentic workflows, memory systems, observability, human handoff.

Phi, Qwen, Gemma; on-device inference; distillation; tool-call fine-tunes.

INT8/INT4, GGUF, AWQ, GPTQ, SmoothQuant, FP8 KV cache.

Model Context Protocol — server design, OAuth 2.1, transports, testing patterns.

Agent Development Kit — lifecycle, tools, evals, guardrails, sessions, skills.

Complete coverage: runtime, model package (BaseLLM + Gemini), persistence, memory, tools, orchestration, A2A transport, observability, evaluation, safety, deployment. All with Java code.

DP, trees, graphs, strings, geometry, network flow, number theory, plus cryptography (AES/RSA/ECC/Ed25519/SHA/HMAC/Argon2/ChaCha/Kyber/ZK-SNARKs/MPC/FHE/JWT/OAuth) and ML (regression/RF/XGBoost/SVM/k-means/PCA/attention/transformer/CNN/RNN/BERT/GPT/Q-learning/PPO/RLHF) — all with animated visualizations.

Zero/few-shot, CoT, self-consistency, ToT, ReAct, self-refine, Reflexion, Constitutional AI, DSPy, meta-prompting, RAG (HyDE/GraphRAG/contextual), agentic prompting, multi-agent orchestration, function calling, structured outputs, prompt caching, cost optimization, A/B testing, eval frameworks.

OWASP LLM Top 10, prompt injection (direct/indirect/multi-turn/GCG), jailbreaks, data exfil, model stealing, MITRE ATLAS, NeMo/Guardrails AI/Llama Guard, PII redaction, hallucination detection, sandboxing, red team (PyRIT/Garak), NIST RMF, EU AI Act, mechanistic interpretability, SAE, watermarking, PQ migration, agent security.

Full protocol coverage: JSON-RPC flow, task lifecycle, streaming, push notifications, artifacts, auth, tracing, MCP interop.

Agent Payment Protocol — subscriptions, settlement, chargebacks, compliance.

Data modeling, consistency, compaction, repair, vector search 5, multi-DC.

Raft, consensus, CRDTs, vector clocks, BFT, quorum systems, 2PC.

LinkedIn, YouTube, TikTok, Airbnb, Amazon, Reddit, Spotify, DoorDash, Grab plus WhatsApp/Signal/Stripe/S3/Google Search/Zoom + core patterns — all with architecture diagrams.

Postgres internals, isolation levels, replication, pgvector, DuckDB, CDC.

Kafka, Flink, Pulsar, exactly-once, CDC, tiered storage, backpressure.

Virtual threads, executor pools, fork/join, actor model, work-stealing, coroutines — each with an architecture diagram.

OpenTelemetry, Prometheus at scale, SLOs, eBPF, alert fatigue.

mTLS, OAuth+PKCE, passkeys, SBOM, zero trust, K8s pod security.

TLS 1.3, DNS, BBR, anycast, mTLS rotation, service mesh patterns.

gRPC, WebSocket, HTTP/2, HTTP/3, SSE, reconnect strategies, HOL blocking.

Opus, jitter buffers, VAD, AEC, LUFS targets, WebRTC pipeline.

HLS, DASH, AV1, ABR strategies, bitrate ladders, DRM, VMAF.

How to design APIs, run postmortems, estimate, pick brokers, review PRs.

Miscellaneous engineering topics that don't fit cleanly elsewhere.