Tag: #series
Writing on GPUs, LLMs, MLOps, Kubernetes — and mindset · 13 posts
Voice AI in Practice — The Complete Guide: Real-Time STT/TTS, Speech LLMs, Turn-Taking, and Deepfake Defense (2025)
Why "AI without a screen" became the hottest product category of 2025. Real-time voice pipelines (VAD/STT/LLM/TTS), speech LLMs (GPT-4o realtime/Gemini Live/Moshi), turn-taking and interruption, emotion and prosody contr
2026-04-15 · 12 min read #voice-ai#stt#tts#gpt-4o-realtime#moshiObservability 2025 Complete Guide: OpenTelemetry, Grafana / Datadog / Honeycomb / SigNoz, SLO and Error Budget, LLM Observability (2025)
Season 5 Ep 8. Without observability there is no operations, and without operations there is no product. Unifying the three OpenTelemetry signals (metric, log, trace), the Grafana stack (Prometheus, Loki, Tempo, Mimir) v
2026-04-15 · 11 min read #observability#opentelemetry#grafana#datadog#honeycombThe Complete LLMOps Guide: Three-Axis Versioning of Model, Prompt, and Eval Set, Canary, Cost Control, and the Platform Team (2025)
Building an LLM product fast has become easy. Running it sustainably is the hard part. Three-axis versioning across model, prompt, and eval set; Shadow/Canary/Blue-Green deployment; cost control through tokens, caching,
2026-04-15 · 11 min read #llmops#mlops#devops#canary#cost-controlFeature Store and the Vector·Graph·Time-Series DB Convergence Guide: Feast·Tecton·Pinecone·Weaviate·Milvus·Neo4j·TimescaleDB (2025)
Season 5 Ep 6. The language of ML is features, the language of AI is vectors, the language of relationships is graphs, and the language of operations is time series. In 2025 every one of these DB categories crosses its o
2026-04-15 · 12 min read #feature-store#vector-db#graph-db#timeseries-db#feastStreaming vs Batch Redefined: Flink, RisingWave, Materialize, CDC, Streaming SQL and the Pragmatism of Real Time (2025)
Season 5 Ep 2. "Everything in real time" was the promise of 2020. 2025 is the era of pragmatism — the line between batch and streaming has blurred, and the real question has become "how fresh does this data have to be to
2026-04-15 · 13 min read #streaming#batch#flink#risingwave#materializeSemantic Layer·Metrics Store·Reverse ETL, The Complete Guide: dbt·Cube·Looker·Hightouch·Census, and Data Activation (2025)
Season 5 Ep 5. If the pipeline produces the data, the Semantic Layer speaks its meaning. The lineage of dbt Semantic Layer, Cube, AtScale, and Looker; the Metrics Store pattern; Headless BI (Lightdash, Transform); the fl
2026-04-15 · 11 min read #semantic-layer#metrics-store#reverse-etl#headless-bi#dbt-semantic-layerData Governance, Lineage and PII Complete Guide: OpenLineage, Collibra, Atlan, DataHub, Unity, Polaris, GDPR, Korean PIPA (2025)
Season 5 Ep 7. The more data you accumulate, the murkier it becomes who uses it, where, why and how much. The OpenLineage standard, the four major data catalogs (Collibra, Atlan, DataHub, Alation), governance in Unity Ca
2026-04-15 · 13 min read #data-governance#lineage#pii#openlineage#collibraThe Complete Guide to LLM Security: Prompt Injection, Jailbreak, Red Team, OWASP LLM Top 10, EU AI Act (2025)
Across 2024–2025 a "security incident" is always among the TOP 3 causes of LLM product failure. 12 prompt injection variants, jailbreak techniques, data exfiltration, model extraction, red team automation (PyRIT/Garak),
2026-04-15 · 12 min read #llm-security#prompt-injection#jailbreak#red-team#owasp-llmThe Complete Guide to Multimodal LLMs: Vision, Document Understanding, OCR, Video, Audio, and the Specifics of Korean (2025)
The text-only era is over. In 2025, LLMs handle images, documents, video, and audio naturally. GPT-4o/Claude 3.5/Gemini/Qwen2-VL/Pixtral compared, Document AI and layout understanding, the modernization of OCR, video and
2026-04-15 · 14 min read #multimodal#vision-llm#document-ai#ocr#whisperOLAP Engines 2025 Comparison Guide: DuckDB, ClickHouse, Snowflake, StarRocks, Pinot, Druid, Trino, Benchmark Traps, Engine Placement (2025)
Season 5 Ep 3. No single engine covers all of OLAP. DuckDB brings the single-node revolution, ClickHouse handles real-time OLAP, Snowflake and BigQuery offer managed convenience, StarRocks, Doris, Pinot and Druid take on
2026-04-15 · 12 min read #olap#duckdb#clickhouse#snowflake#bigqueryLLM Evaluation & Observability: Eval Harness, LLM-as-Judge, Tracing, Regression Prevention (2025)
Claiming a new model "seems better" is not science. The 1 failure mode of 2025 LLM products is "not measuring". Eval harness design, LLM-as-judge pitfalls and calibration, 3-layer observability (Trace/Span/Metric), produ
2026-04-15 · 11 min read #llm-eval#observability#llm-judge#distributed-tracing#langsmithData Orchestration 2025 Complete Guide: dbt, SQLMesh, Dagster, Airflow, Prefect, Data Contracts, CI/CD (2025)
Season 5 Ep 4. If engines query, orchestrators govern the pipelines. The standardization of dbt and its limits, the revolt of SQLMesh, the asset-centric model of Dagster, the evolution of Airflow 2.x and 3.0, the rebirth
2026-04-15 · 13 min read #data-orchestration#dbt#sqlmesh#dagster#airflowAI Product Design Complete Guide: Generative UI, Trust, Feedback, Streaming, Agent UX, Korean Culture Specialization (2025)
A good AI product is not "a good model + good UX" — it is "UX design that builds trust within constraints." The boundary between deterministic and generative UI, designing for trust, uncertainty, and editability, streami
2026-04-15 · 10 min read #ai-product-design#ux#generative-ui#trust#streaming