Entity-based discovery,
built around four workstreams.

e.g. Tech Lead, Product Lead, VP Engineering, Director of Applied AI.

Tech lead, product lead, eng manager, exec sponsor, etc. Name + role + how they want to be engaged.

Single name preferred — the person whose "yes" moves us forward.

e.g. platform, data infra, security, applied AI — we'd love to map the touch points.

Customer demand, competitive pressure, platform consolidation, regulatory, internal mandate — what's the forcing function?

Helpful to know what worked, what didn't, and what we shouldn't re-litigate.

Pretend you're describing it to a new engineer joining the team. We want your framing, not ours.

What can't users do, or what do they do badly, that this initiative fixes?

e.g. "find every email between Person A and any external party at Domain X about Topic Y between dates." Concrete examples = sharper backend.

Adoption, search precision, time-to-answer, cost per matter, customer NPS, engineering productivity — whatever matters most to you.

What precision floor will the downstream UX accept?

Internal labelled datasets, prior model accuracy, manual review timings — anything we can anchor to.

Defensibility, explainability, auditability, latency, cost. Especially relevant for litigation-grade outputs — defensibility often outranks raw precision.

Drives parsing complexity. Rough percentages are fine.

Affects segmentation cues — quoted-reply markers vary by language and client.

Is it always a single email, or sometimes a thread, or sometimes a reply within a quoted chain? The unit definition is the spec.

Forwarded chains, embedded replies, inline annotations, mixed-language quoting, etc. The pathological 5–10% usually determines production readiness.

Treated as part of the parent message, separate units, or linked entities?

Tools, models, vendors, in-house code — what's the baseline we're improving on?

If you have a point of view, share it. If you don't, we'll bring one.

cc, bcc, return-path, reply-to, signature phone numbers, signature URLs, titles, company names, departments, disclaimers, etc.

Different reliability profiles. Affects how we weight confidence.

Templated corporate, free-form personal, mixed? A few sample sanitized signatures would be 10× more useful than any answer here.

Forwarded-chain pollution is a common failure mode — worth flagging if it matters.

High-leverage feature for the resolution layer and for defensibility. Most modern pipelines do this.

More attributes = stronger downstream resolution but more work to QA.

If you have a schema, share it. If not, we'll propose one.

Corporate directories, customer-supplied custodian lists, public sources.

Aggressive collapsing reduces dupes but risks false merges. We can tune per use-case.

Human-in-the-loop is often the right answer for litigation-grade workflows.

Just the email corpus, or also chat, documents, HR data, custodian lists, third-party sources? Cross-corpus linking is often the highest-value use case.

e.g. same email address used by two real people over time. The hardest part of identity.

Native graph DB (Neo4j, Neptune), relational with adjacency tables, search index, or vector store. Trade-offs differ.

Lookup by ID, fuzzy lookup, traversal, batch enrichment — the API shape drives a lot of internal design.

e.g. "this email belonged to Person A from 2018–2022, then Person B." Litigation-grade graphs often need this — a significant design decision if yes.

Often non-negotiable for legal defensibility. Affects model choice (interpretable vs. black-box).

Internal eDiscovery platform, document review tools, ingestion pipelines, identity / auth, billing, telemetry.

A Figma share link is ideal. Static images work too — we'll just need a walk-through.

Vendor, on-prem only, no third-party APIs, specific approved models. Common constraint in legaltech given client confidentiality requirements — worth knowing day one.

Per-matter isolation, regional residency, dedicated vs. shared clusters, customer-managed keys, etc.

Managed Engineers = ISA-led pod with a tech lead, outcomes-oriented. Staff Aug = engineers embedded into your team day-to-day.

Typical candidates: Tech Lead, Backend Engineer, ML/NLP Engineer, Data Engineer, DevOps/SRE, QA, Solutions Architect, Project Manager.

e.g. you'll keep product, design, and DevOps in-house — useful so we don't over-staff.

e.g. signature-extraction experience, identity-resolution at scale, eDiscovery domain depth, specific LLM tooling.

Standups, demos, planning, on-call expectations. Share what's worked best with past partners.

ISA's senior team is in Kosovo and Albania (EU hours, strong US overlap). Helpful to know what to lean into and what to avoid.

Customer commitments, regulatory dates, board demos, RFP timing.

Helps us right-size the proposal. Approximate is fine.

Technical, organizational, commercial, regulatory — whatever's on your mind.