Divyajeet

01

Turns plain-English analytics questions about the Olist e-commerce dataset into Cypher and runs them against Neo4j.

What I built

• Query generation from the graph schema with a local Ollama model
• A validator, an EXPLAIN check and a read-only account on every query

Matches 91 of 112 verified benchmark questions (81%)

How it flows

1. Plain-English question
2. Local Ollama modelwrites Cypher from the graph schema
3. Validator + EXPLAINchecked before it runs
4. Neo4jread-only account
5. Answer

Try a question

MATCH (o:Order {status: 'delivered'}) RETURN count(o) AS deliveredOrders

MATCH (:Order {status: 'delivered'})-[:HAS_ITEM]->(i:OrderItem)
      -[:OF_PRODUCT]->(:Product)-[:IN_CATEGORY]->(c:Category)
RETURN c.nameEnglish AS category,
       round(sum(i.price) * 100) / 100 AS deliveredItemGmvBrl
ORDER BY deliveredItemGmvBrl DESC, category
LIMIT 5

MATCH (o:Order {status: 'delivered'})-[:HAS_REVIEW]->(r:Review)
WHERE o.deliveredAt > o.estimatedDeliveryAt
RETURN avg(r.score) AS averageReviewScore

MATCH (c:Customer)
RETURN c.state AS state, count(DISTINCT c.uniqueCustomerId) AS customers
ORDER BY customers DESC, state
LIMIT 5

Olist Query StudioPlain-English questions to Cypher

02

Turns free-text bug reports into schema-validated triage records, from category and severity to a suggested next step.

What I built

• A FastAPI service running a local Qwen 2.5 model
• Guardrails that mark anything the report doesn't support as “unknown”
• A React review UI that keeps the JSON in view

30 of 30 benchmark responses schema-valid

How it flows

1. Free-text bug report
2. FastAPI + Qwen 2.5runs locally
3. Guardrailsunsupported claims become “unknown”
4. Schema check
5. React review UI

Try a report

{
  "category": "ui",
  "severity": "low",
  "priority": "P4",
  "environment": "Safari",
  "reproducible": true,
  "probable_cause": "The dark mode setting may not update the background color of the settings panel.",
  "suggested_action": "Verify that the dark mode setting correctly changes the background color of the settings panel."
}

{
  "category": "database",
  "severity": "critical",
  "priority": "P1",
  "environment": "unknown",
  "reproducible": "unknown",
  "probable_cause": "The migration script may have used a destructive DELETE statement instead of an upsert or a restore function.",
  "suggested_action": "Review the migration script to ensure it uses the correct SQL commands and includes a rollback plan if necessary."
}

{
  "category": "database",
  "severity": "medium",
  "priority": "P3",
  "environment": "unknown",
  "reproducible": true,
  "probable_cause": "The save operation may not update the record ID correctly.",
  "suggested_action": "Verify the order note is being saved with a unique ID each time."
}

{
  "category": "security",
  "severity": "high",
  "priority": "P1",
  "environment": "unknown",
  "reproducible": "unknown",
  "probable_cause": "The URL parameter may bypass authentication checks.",
  "suggested_action": "Verify that the invoice ID is authenticated before accessing the invoice content."
}

BugSieveBug reports to validated JSON

03

A command-line tool I use daily that files every loose download by category, extension and month.

What I built

• Filing rules by category, extension and month
• Exact-duplicate detection by size and MD5, sent to the Recycle Bin
• A colour-coded preview of every move before it runs

Parity test: files everything exactly as the PowerShell original

How it flows

1. Downloads folder
2. Sort rulescategory, extension, month
3. Duplicate checksame size and MD5
4. Colour-coded preview
5. Moveduplicates to the Recycle Bin

Sort-DownloadsPython CLI with a rich preview

04

Compares PDF/DOCX resumes with job descriptions and returns ATS-style feedback, keyword gaps and suggestions.

What I built

• A Next.js frontend and an Express/TypeScript API that share Zod schemas
• Supabase auth
• OpenAI-assisted feedback backed by deterministic scoring

In progress · repository not public yet

How it flows

1. Resume + job descriptionPDF or DOCX
2. Next.js UISupabase auth
3. Express/TypeScript APIZod schemas shared with the UI
4. Scoringdeterministic, with OpenAI-assisted feedback
5. ATS-style report

05

A local system-monitoring dashboard for CPU, RAM, GPU, storage, thermal, network and process telemetry.

What I built

• An Express backend that polls telemetry behind cached endpoints
• A React/Vite UI with live cards, time-series charts and process tables

In progress · repository not public yet

How it flows

1. System telemetryCPU, RAM, GPU, storage, thermal, network, processes
2. Express poller
3. Cached endpoints
4. React/Vite UIlive cards, charts, process tables