Meetings, email, chat and filled-in forms
Extraction from documents with no fixed layout at all — threads, transcripts, survey text and handwriting.
Everything else people call document extraction has a document. An invoice has a total somewhere, a passport has a machine-readable zone in a fixed place, a tax form has numbered boxes that mean what the authority says they mean. You can be wrong about where a field is, but the field exists at a coordinate and you can check yourself against the form. None of that is true here. Minutes, an email export, a chat log and a free-text survey answer have no layout to learn, and the reason they are grouped together is that each one has to find its structure somewhere other than the page.
Usually it is there. Email carries a real parent-child graph in its headers, so thread order is a lookup rather than a guess. Minutes are an edited document with numbered items and a convention about which verbs record a decision. A chat export has an opaque speaker id and a timestamp encoding you can read from the vendor’s own format notes. Survey text has nothing at all, which is why the codebook you write becomes the structure and the deliverable both. These pages are about finding that hidden structure and then about the specific ways each one lies to you.
Extracting Action Items and Owners From Meeting Minutes
A schema and a worked run for pulling owner, task and due date out of minutes that already exist, including the duplicate-action and carry-forward traps.
10 min read
Extracting Attendee Lists From Meeting Minutes
Why present, in attendance and apologies are three different states, and what collapsing them into one list of names destroys.
9 min read
Extracting Decisions Made From a Set of Meeting Minutes
The wording cues that separate a recorded decision from a discussion point that sounds like one, and why the verb is more reliable than the model's judgement.
9 min read
Extracting Sender, Recipients and Thread Structure From an Email Export
Reconstructing a real reply tree from RFC 5322 References and In-Reply-To headers, and separating quoted history from new text.
11 min read
Extracting Action Items From an Email Thread
Pulling an owner-task-deadline record out of thread bodies, and what to do with the commitment whose owner is genuinely not determinable.
10 min read
Extracting Attachments Metadata From an Email Export
Reading filename, size and type out of MIME parts, telling a real attachment from an inline signature logo, and catching the attachment a message references but does not contain.
10 min read
Extracting Speaker Turns From a Chat Transcript
How support-widget, SMS and messaging-app exports each encode who is speaking, and a normalisation schema that survives all three.
10 min read
Extracting Timestamps From a Chat Transcript With Inconsistent Formats
A transcript mixing '2m ago' with 'Yesterday 3:14 PM' and full ISO stamps, and a method that resolves all three to one zone-aware instant without inventing any.
10 min read
Extracting Structured Fields From a Support Ticket Body
A schema for product, version, severity, steps to reproduce and expected-versus-actual, built so the fields a ticket does not contain stay empty.
11 min read
Extracting Order Numbers Referenced Inside a Support Ticket
Candidate generation for the several shapes a customer pastes an order number in, then resolution against your own order table rather than by pattern alone.
9 min read
Extracting Coded Answers From Open-Ended Survey Responses
Building a fixed codebook, applying it to free-text answers, and measuring agreement per code — including why a rare code always looks terrible.
12 min read
Extracting Numeric Ratings Embedded in Free-Text Survey Comments
Reading a stated rating out of a comment, keeping it in its own column, and why the scale answer must never be overwritten by it.
9 min read
Extracting Checked and Unchecked Boxes From a Scanned Form
Why proving a box is empty takes two detections rather than one, and a threshold policy for pencil marks, crosses, folds and bleed-through.
11 min read
Designing a Schema for a Form With Optional and Repeating Checkbox Groups
Modelling check-all-that-apply as a set and pick-one as an enum, with the empty-versus-absent distinction and a representation for states the form forbids.
10 min read
Extracting Text From Handwritten Meeting Notes
The conventions meeting notes carry that general handwriting does not — margin owners, arrows, ditto marks and a personal abbreviation set — and why the transcription must stay verbatim.
10 min read
Extracting Structured Fields From a Handwritten Order Form
Quantity and product code on a handwritten order, where a product code has a check digit to catch a misread and a quantity has nothing at all.
10 min read
Other topics
- LLM fundamentals & architecture
- Tokens, tokenization & context windows
- Prompt engineering
- Reasoning models & test-time compute
- Multimodal AI: vision, audio, video
- RAG & retrieval
- Embeddings & vector search
- AI agents & tool use
- Structured output & function calling
- Fine-tuning & post-training
- Local inference errors, string by string
- Running local models day to day
- Testing code that calls an LLM
- Snapshot and property testing for model output
- Regression suites for prompts
- Eval gates in CI
- Flaky tests against a model
- Determinism and the cost of testing
- Contract and streaming tests
- Testing tool calls and retrieval
- Inference, serving & latency
- Rolling out a prompt change
- Testing AI systems in practice
- Forecasting a time series
- Machine learning on tabular data
- Geospatial data and models
- Understanding audio that is not speech
- Understanding video
- Core computer vision tasks
- Machine learning on graphs
- Point clouds and 3D
- Evaluation, benchmarks & LLM-as-judge
- Sensor and IoT data
- Logs and event streams
- Models over biological sequences
- Machine learning on molecules
- Embedding and searching code
- Extracting invoices and purchase orders
- Receipts, statements and tax forms
- Insurance policies and contracts
- Deeds, court filings and patents
- Extracting from medical records
- Observability & LLMOps
- CVs, certificates and identity documents
- Shipping, customs and technical documents
- Building an extraction pipeline
- Business, property and inspection documents
- Contract clauses and insurance claims
- Regulated and compliance documents
- Consumer, travel and closing documents
- Mapping one chat API onto another
- SDK and framework migrations
- Hallucination & failure modes
- Re-embedding and model deprecation
- Cutting over between providers
- Parity gaps, shims and legacy endpoints
- Moving between model versions
- Migrating vector stores and caches
- Mapping capabilities and parameters
- Migrating pipelines and agents
- Contracts, runbooks and rollback
- Auditing a codebase before a cutover
- Compliance and fine-tune migration
- LLM cost engineering
- Routing, cost tracking and multi-tenancy
- What a migration does to your prompts
- AI security & prompt injection
- Privacy, compliance & data residency
- AI governance, policy & society
- Building reliable AI applications
- AI hardware, GPUs & compute
- Open-weight models & local inference
- AI for developers & coding agents
- AI in industry: vertical playbooks
- AGI, superintelligence, alignment & the long future
- Machine learning foundations
- NLP fundamentals & classical tasks
- Data engineering for AI
- Synthetic data & dataset curation
- AI product design & UX
- Search, ranking & recommendation
- Enterprise adoption & change management
- AI careers, skills & teams
- Reading AI research
- AI in science & discovery
- Robotics & embodied AI
- AI economics, markets & business models
- AI myths, hype & media literacy
- Context engineering
- Shipping AI features: patterns & anti-patterns
- Build it: end-to-end AI tutorials
- Python for AI: hands-on recipes
- TypeScript, React and the web
- Frameworks and SDKs
- Errors and troubleshooting
- AI facts, numbers and statistics
- The history of AI
- The maths behind AI
- Architectures beyond the transformer
- Reinforcement learning
- Diffusion and generative media
- Speech, audio and voice engineering
- Benchmarks, one at a time
- AI search visibility
- Infrastructure and operations
- Databases and storage for AI
- Knowledge graphs and structured knowledge
- Classical ML in production
- Regulation, jurisdiction by jurisdiction
- Prompt recipes and pattern library
- AI for people who do not write code
- Writing, media and creative work
- Edge and on-device AI
- Interpretability and model internals
- Field notes
- OpenAI model behaviour
- Claude model behaviour
- Gemini model behaviour
- Llama model behaviour
- Mistral model behaviour
- Qwen model behaviour
- DeepSeek model behaviour
- Cohere model behaviour
- Grok model behaviour
- Small model behaviour
- Hybrid model architectures
- Token cost by language and script
- Transliteration, romanization and script handling
- Locale-correct output
- Multilingual generation quality
- Multilingual pipelines
- The EU AI Act, article by article
- AI under the GDPR and EU data law
- US AI regulation, state and sector
- International AI governance and standards
- AI litigation and enforcement
- Running AI workloads on AWS
- Running AI workloads on Google Cloud
- Running AI workloads on Azure
- AI at the edge: Workers, Vercel and Netlify
- Serving models on Kubernetes
- Operating AI infrastructure
- Quantization formats and what they cost
- llama.cpp, flag by flag
- Ollama and the desktop local-model runtimes
- Local models on Apple Silicon
- Hardware for local inference
- Running speech and embedding models locally
- Model files, adapters and conversion
- VRAM arithmetic for local models