<- Featured work
FEATURED / SYNTHETIC-DATA PROTOTYPE

PlantBrain

A team-built knowledge-intelligence prototype that grounds operational answers in source evidence.

Next.jsTypeScriptSQLiteBM25Cited retrieval
GitHub repo (aj branch)
The system

Operational troubleshooting needs answers that make their supporting evidence easy to inspect.

Collaborative AI product prototype for cited troubleshooting and operational learning.

Project visuals

See the system before the explanation.

Diagrams and project artifacts, each shown with its current evidence status.

Diagram showing synthetic operational records indexed with SQLite and BM25, a question retrieving evidence, and a cited answer with an extractive fallback when an LLM is unavailable.
Cited evidence flow Based on the reviewed aj-branch project summary; all demonstration records are synthetic.
01 / Build

What exists

Project scope

  • A collaborative prototype explores cited answers for troubleshooting, root-cause investigation, compliance gaps, and lessons learned.
  • The reviewed local demonstration retrieves from synthetic operational records with SQLite/BM25.
  • When an LLM is unavailable, the demonstration can return an extractive answer with citations.

Architecture

  • Synthetic records form the local demonstration corpus.
  • A question retrieves supporting passages through SQLite/BM25 and keeps source citations with the response.
  • The response path can use an extractive cited fallback when an LLM is unavailable.
02 / Ownership

Product & Frontend Co-Developer

Co-developed the Next.js operations interface and evidence views, integrated the front and back ends, and worked on evaluation monitoring and evidence integrity.

Contribution

  • Built product UI and document evidence views for cited troubleshooting flows.
  • Worked on frontend/backend integration, evaluation monitoring, and stale-result handling.
  • Added backend evidence-integrity guards so responses stay tied to their source material.

Collaborative scope

  • Co-developed with Mridul Negi using synthetic operational records.
  • Mridul led ingestion, retrieval, graph, and agent work; this page does not attribute those areas to AJ.
03 / Decisions

Engineering judgment

Key decisions

  • Keep the source document visible beside the answer.
  • Treat stale results and evaluation state as product behavior that needs explicit handling.
  • Preserve an extractive cited path when model output is unavailable.

Proof highlights

  • Owner-reviewed project and contribution wording identifies the aj branch and synthetic-data boundary.
  • The branch ref and repository link were checked; the local checkout was not used as evidence for the newer reviewed state.
Product workflow

From discovery to coordination.

Question

Ask about a troubleshooting, cause, compliance gap, or lesson.

Retrieve

Find supporting passages in the synthetic local SQLite/BM25 demonstration.

Cite

Keep document evidence visible with the answer, including the extractive fallback path.

Boundary

What this evidence does not claim

The reviewed demonstration uses synthetic records; deployment, customer, and outcome claims are not part of this case study.

  • All cited demonstration records are synthetic; no industrial deployment or customer impact is claimed.
  • No accuracy, latency, adoption, or time-saved result is presented.
  • Retrieval, graph, ingestion, and agent work is credited to the teammate who led those areas.