Changelog
What's new in Lookato
Product updates, new features, and improvements shipped every week. We build in the open.
May 2026
Lookato G&J — Operations intelligence for the Gems & Jewellery industry
We've focused Lookato on what we do best: helping diamond and jewellery manufacturers run their operations with real-time visibility, governed metrics, and proof for every answer. New dedicated product page, 5 industry-specific use case pages, events page for IIJS Mumbai, and a complete website repositioning for the G&J industry.
NewLookato G&J product page — detailed walkthrough of conversational queries, evidence pane, governed metrics, pre-built dashboards, and ERP integration for diamond and jewellery operations.
New5 industry-specific use case pages: diamond manufacturing, jewellery manufacturing, diamond trading, lab-grown diamond, and integrated houses. Each with real operational questions and dashboard previews.
NewAbout page — our decade in Surat building Trybe, Ornix, and DiaDNA for the Gems & Jewellery industry.
NewEvents page — meet us at IIJS Mumbai (August 2026), Hong Kong Jewellery Fair (September), and Vicenzaoro (September).
NewQualified demo request form with G&J-specific fields: business type, current ERP systems, company size.
ImprovedComplete website repositioning: homepage, features, how-it-works, pricing, security, navigation, and footer — all rewritten for Gems & Jewellery operations.
May 8, 2026
Governed Analytics, Evidence Pane & Semantic CI
The trust update. Every answer now carries provenance — which governed metric was used, which formula was applied, and whether the plan was validated. Plus golden question regression suites that catch governance drift before users do.
NewGoverned Metrics Layer — define metrics with exact formulas (definition_json), compiled deterministically to LQL expressions. Three-step enforcement: resolve metric from query, inject into planner prompt, validate the generated plan. Published metrics are versioned with approval workflow.
NewEvidence Pane — every answer shows a collapsible provenance panel: governed metric name + version, formula used, dataset, freshness, match type (exact/alias/semantic), and validation status. Business View for all users; Analyst View (SQL, compiled expression, LQL plan) for admin/analyst roles. Analyst-only fields redacted server-side.
NewGolden Questions + Semantic CI — register golden questions with approved baselines. Run on-demand regression suites that compare responses against baselines. Drift comparator checks: dataset, governed metric presence (by ID), compiled formula, dimensions, filters, query path. Catches governance drift from prompt changes, model upgrades, or glossary edits.
NewDeterministic metric compiler — definition_json compiles to dialect-neutral LQL expressions (ratio, aggregation, filtered_count). The LLM selects the metric; the compiler constructs the exact formula. No hand-authored governed calculations.
NewGoverned metric ambiguity clarification — when multiple governed metrics match (e.g., "margin" matches both Gross Margin % and Net Margin), the system presents clickable pill buttons for disambiguation instead of guessing.
NewAdmin metrics management page — CRUD for governed metrics with versioning, aliases, status lifecycle (draft → staging → published → archived), and audit logging. Admin golden questions page with baseline review, suite execution, and drift results.
New15 seed golden questions for diamond trading — including adversarial synonym tests ("What is profit margin?" → should resolve to governed Gross Margin %). Slug-to-ID resolution ensures stable test references.
ImprovedEvidence persisted with every assistant message — reopening conversations renders stored evidence snapshot, not regenerated data. Evidence survives across sessions.
FixedFixed analytical orchestrator fallback: governed metric validation failures and ambiguity clarifications no longer silently fall through to single-shot path.
FixedFixed ClarificationCard styling for dark theme — replaced CSS variable styling with Tailwind classes for reliable rendering.
April 24, 2026
Conversational Intelligence, Admin Glossary & Exports
The biggest update yet. A multi-layer AI intelligence system that understands your business terms, asks when unsure, and guides exploration. Plus admin glossary management and CSV exports.
NewConversational Intelligence — multi-layer intent resolution with nano classification (GPT-5.4-nano), full resolution (GPT-5.4-mini), and binding constraints to the query planner
NewSmart Clarification — ClarificationCard UI with clickable options when terms are ambiguous. Structured API responses for reliable selection.
NewBusiness Glossary with ambiguity flags — define ambiguous terms with multiple interpretations. Managed via admin UI, no SQL required.
NewSmart follow-ups — data-aware suggestion chips generated from actual query results, not generic prompts
NewCoaching banner — progressive reveal showing intent, confidence, and assumptions while you wait for results
NewPost-execution LLM narrative — GPT-5.4-mini generates contextual narratives in parallel with deterministic fallback
NewPDF sharing with conversation context — prior Q&A pairs shown above the shared answer for full context
NewAdmin glossary management page — table view, add/edit modal with translation templates, delete confirmation
NewAdmin exports page — CSV export with sync download (< 10K rows) or async job tracking with polling
NewDomain guidance — enriched dataset domainContext in planner prompt replaces YAML playbooks
ImprovedGPT-5.4 model upgrade — centralized model config with env var overrides, all LLM call sites migrated
ImprovedQuery latency optimization — removed redundant LLM call (2-5s savings), parallelized orchestrator setup, eliminated duplicate data fetches, granular SSE status updates
ImprovedIntent resolver parser made tolerant of LLM output variations — accepts missing optional fields, coerces types
FixedFuzzy resolver below-threshold bug — candidates from ClickHouse that scored below threshold now trigger clarification instead of silent no-match
FixedInvalid resolvedMetrics/breakdownDimensions filtered against dataset schema before reaching planner
FixedcountIf without field auto-corrected to count instead of validation error
FixedPie/donut chart blocked for single-category data — falls through to bar chart
FixedClarification response JSON no longer shows as raw text in chat bubbles or exported PDFs
March 31, 2026
Live Dashboards, Chat-to-Notebook & PDF Sharing
Pin insights to live dashboards, save analyses to notebooks, and share any answer as a beautifully formatted PDF.
NewLive dashboards — pin any chat answer to a dashboard card with auto-refresh
NewChat-to-notebook — save any analysis to a collaborative notebook with one click
NewPer-answer PDF export — beautifully formatted A4 PDF with header, narrative, charts, tables, and metrics
ImprovedQuery inspector — see the SQL, execution stats, and step details behind any answer
March 23, 2026
Watches & Alerts System
Monitor any metric with scheduled watches. Get notified via Slack or email when thresholds are crossed.
NewWatches — schedule recurring queries with threshold alerts
NewAlert delivery via Slack and email with investigation links
ImprovedWatch management page with status tracking, mute/unmute, and history
March 19, 2026
Premium Chat UX & Visual Polish
Redesigned chat experience with premium pipeline rendering, key insights, confidence indicators, and follow-up chips.
NewPremium pipeline — structured rendering with key insight headline, confidence badge, metric cards, charts, tables, follow-up chips
ImprovedTypography and visual hierarchy overhaul across the chat interface
See Lookato G&J in action with your data.
Book a 30-minute demo with someone who knows the Gems & Jewellery industry.
By the makers of Trybe and Ornix · Founded in Surat, the world's diamond capital