metabase/metabase
Mar 2026 – Apr 2026
2 Months active
Languages Used
ClojureSelmer
Technical Skills
AI integrationClojureSQLfull stack developmentAPI developmentbackend development
Mihael Konjević
PROFILE
Mihael Konjević
During two months on the metabase/metabase repository, Konjevic developed and delivered MetaBot v3, a native Clojure agent that replaced an external Python AI service with an in-process, low-latency implementation. He introduced a robust prompt system supporting multiple SQL dialects, context enrichment, and direct Toucan2 database access, enabling safer and more efficient LLM-driven analytics workflows. Konjevic also aligned NLQ and Metabot profiles for consistent user experience and built agent-lib for structured MBQL processing, including program repair, validation, and evaluation. His work demonstrated deep expertise in Clojure, SQL, and backend development, with comprehensive test coverage and improved observability.
Overall Statistics
Feature vs Bugs
100%Features
Repository Contributions
3Total
Bugs
0
Commits
3
Features
3
Lines of code
46,107
Activity Months2
Work History
April 2026
2 Commits • 2 Features
Apr 1, 2026April 2026 (2026-04) monthly summary for Metabot work: Key features delivered: - NLQ Metabot profile alignment and testing: aligned embedding profile with NLQ profile (same system prompt and tool set), removed obsolete test, and added a new test to verify alignment, ensuring consistent NLQ/metabot user-facing behavior. - Structured MBQL processing with agent-lib: introduced a full agent-lib module for structured MBQL program repair, validation, and evaluation; provides idempotent normalization, schema validation, and safe evaluation; replaces old query construction with a unified program format and enhances querying capabilities. Major bugs fixed: - Fixed evaluation context and time-interval keyword coercion; aligned source type mappings and IDs to enable correct evaluation; addressed test failures and updated to the new program schema. - Stabilized tests and build tooling: fixed tests, addressed Eastwood warnings, updated cljfmt indentation, and refined prompt guidance to reflect new tool schema and field IDs. Overall impact and accomplishments: - Significantly improved NLQ/metabot consistency and user experience, reducing edge-case behavior and support needs. - Enabled richer, safer MBQL queries via agent-lib, improving reliability, maintainability, and future extensibility of Metabase’s AI-assisted querying. - API and tooling improvements (v2 endpoints, MCP-friendly schemas) open path for broader client adoption and automation with fewer integration issues. Technologies/skills demonstrated: - Clojure, CLJ tooling (clj-kondo, Eastwood), and codebase modernization - MBQL domain knowledge and agent-lib architecture (repair, validate, eval, runtime) - Sandboxed evaluation, runtime scoping optimizations, and schema evolution - API design, tool prompts engineering, and test-driven development
2 Commits • 2 Features
Apr 1, 2026April 2026 (2026-04) monthly summary for Metabot work: Key features delivered: - NLQ Metabot profile alignment and testing: aligned embedding profile with NLQ profile (same system prompt and tool set), removed obsolete test, and added a new test to verify alignment, ensuring consistent NLQ/metabot user-facing behavior. - Structured MBQL processing with agent-lib: introduced a full agent-lib module for structured MBQL program repair, validation, and evaluation; provides idempotent normalization, schema validation, and safe evaluation; replaces old query construction with a unified program format and enhances querying capabilities. Major bugs fixed: - Fixed evaluation context and time-interval keyword coercion; aligned source type mappings and IDs to enable correct evaluation; addressed test failures and updated to the new program schema. - Stabilized tests and build tooling: fixed tests, addressed Eastwood warnings, updated cljfmt indentation, and refined prompt guidance to reflect new tool schema and field IDs. Overall impact and accomplishments: - Significantly improved NLQ/metabot consistency and user experience, reducing edge-case behavior and support needs. - Enabled richer, safer MBQL queries via agent-lib, improving reliability, maintainability, and future extensibility of Metabase’s AI-assisted querying. - API and tooling improvements (v2 endpoints, MCP-friendly schemas) open path for broader client adoption and automation with fewer integration issues. Technologies/skills demonstrated: - Clojure, CLJ tooling (clj-kondo, Eastwood), and codebase modernization - MBQL domain knowledge and agent-lib architecture (repair, validate, eval, runtime) - Sandboxed evaluation, runtime scoping optimizations, and schema evolution - API design, tool prompts engineering, and test-driven development
April 2026
March 2026
1 Commits • 1 Features
Mar 1, 2026March 2026 monthly summary for metabase/metabase: Delivered MetaBot v3 as an in-process native Clojure agent, replacing the external Python AI service with a production-ready, low-latency in-process implementation. Introduced a robust prompt system, context enrichment, and SQL query tooling, enabling faster, more reliable LLM interactions within a single JVM and direct Toucan2 access. Implemented feature-flag controlled rollout (use-native-agent) for gradual validation and minimized risk. Expanded capabilities with 13 wrapped tools and a comprehensive memory/state pipeline to support richer conversations and tool orchestration. Phase-based progress across prompts/context (Phase 2A), SQL tooling (Phase 2B), resource/link tooling (Phase 2C), chart tooling and profiles (2F), plus read_resource tool integration for resource discovery. Achieved high-quality test coverage (210 assertions across 15 tests; all passing) and improved observability via Prometheus metrics and Snowplow events. Business value includes lower latency, reduced HTTP round-trips, improved data access via Toucan2, and safer, structured SQL edits and query operations. The deliverables position MetaBot for scalable, enterprise-grade AI automation with richer analytics workflows.
March 2026
1 Commits • 1 Features
Mar 1, 2026March 2026 monthly summary for metabase/metabase: Delivered MetaBot v3 as an in-process native Clojure agent, replacing the external Python AI service with a production-ready, low-latency in-process implementation. Introduced a robust prompt system, context enrichment, and SQL query tooling, enabling faster, more reliable LLM interactions within a single JVM and direct Toucan2 access. Implemented feature-flag controlled rollout (use-native-agent) for gradual validation and minimized risk. Expanded capabilities with 13 wrapped tools and a comprehensive memory/state pipeline to support richer conversations and tool orchestration. Phase-based progress across prompts/context (Phase 2A), SQL tooling (Phase 2B), resource/link tooling (Phase 2C), chart tooling and profiles (2F), plus read_resource tool integration for resource discovery. Achieved high-quality test coverage (210 assertions across 15 tests; all passing) and improved observability via Prometheus metrics and Snowplow events. Business value includes lower latency, reduced HTTP round-trips, improved data access via Toucan2, and safer, structured SQL edits and query operations. The deliverables position MetaBot for scalable, enterprise-grade AI automation with richer analytics workflows.
Activity
Loading activity data...
Quality Metrics
Correctness100.0%
Maintainability80.0%
Architecture100.0%
Performance80.0%
AI Usage66.6%
Skills & Technologies
Programming Languages
ClojureSelmer
Technical Skills
AI integrationAPI developmentClojureSQLbackend developmentdata processingfull stack developmentquery optimizationtesting
Repositories Contributed To
Overview of all repositories you've contributed to across your timeline