Work

I’ve shipped three projects worth writing up, and I’ve put my work at Accenture in the same list below. Each entry carries a ledger of sourced numbers and the decisions behind them. The three projects link to a repository and a test suite you can check yourself; Accenture’s numbers come from a private export with no public link, stated as such rather than dressed up the same way.

Case studies

Synapse

A self-hosted AI agent platform running across WhatsApp, Telegram, Discord, and Slack.

Built independently

Methodcount of `def test_` definitions across 260 test files on origin/develop, measured with grep -cE '^\s*(async )?def test_'. origin/develop is 11 commits behind origin/mainDate2026-08-11Sample sizen = 260Sourcegithub.com/UpayanGhosh/Synapse-OSS/tree/develop

tests

Methodgit rev-list --count HEAD on origin/main of a full cloneDate2026-08-11Sourcegithub.com/UpayanGhosh/Synapse-OSS/commits/main

commits

Methodprovider keys registered in _KEY_MAP in workspace/sci_fi_dashboard/llm_router.py: anthropic, openai, gemini, groq, openrouter, mistral, togetherai, xai, cohere, minimax, moonshot, zai, volcengine, huggingface, nvidia_nim, qianfan, deepseek. Routed through litellm.Router with a per-role fallback table; Bedrock, Vertex, Ollama and OAuth paths are additional and not counted hereDate2026-08-11Sourcegithub.com/UpayanGhosh/Synapse-OSS/blob/main/workspace/sci_fi_dashboard/llm_router.py

LLM providers routed

Python
FastAPI
LanceDB
SQLite FTS5
asyncio

tldr skill

Turns any repository into an onboarding site: dependency graphs, directory maps, framework cookbooks.

Built independently

Methodnpm registry trailing-12-month download count for tldr-skill v1.6.0Date2026-08-11Sourceapi.npmjs.org/downloads/point/last-year/tldr-skill

npm downloads, trailing year

Python
Claude API
Cytoscape.js
D3.js

claude-jarvis

A universal intent router for Claude Code, matching fast-path first and falling back to dynamic skill discovery.

Built independently

Methodclaude-jarvis downloads over 2026-04-03 to 2026-05-02, the 30 days following release, from the npm registry daily range API. Peak day 2026-04-06 at 1,222Date2026-08-11Sample sizen = 30Sourceapi.npmjs.org/downloads/range/last-year/claude-jarvis

npm downloads, first 30 days

Methodnpm registry trailing-12-month download count for claude-jarvisDate2026-08-11Sourceapi.npmjs.org/downloads/point/last-year/claude-jarvis

npm downloads, trailing year

Methodnpm registry trailing-30-day download count for claude-jarvis v1.2.12. Manual snapshot. Not fetched at build; refreshed by hand.Date2026-08-11Sourceapi.npmjs.org/downloads/point/last-month/claude-jarvis

npm downloads, trailing 30d

TypeScript
Claude Code Plugin API

Accenture

I resolved tickets and root-caused an approval failure that had already been closed twice before it reached me. I also built two internal tools that took routine questions out of my inbox: a self-service database query agent, and a documentation pipeline spanning two production databases.

Built at Accenture · Associate Software Engineer · October 2024 to August 2026

Methodof the 97 work items ever assigned to me (see accenture-work-items-assigned), count whose state I personally transitioned to Resolved or Closed at least once. Same private Azure DevOps export, no public URLDate2026-08-12Sample sizen = 97

work items resolved or closed

Methodpull requests I authored, 2025-02-24 to 2026-08-05, whose state is Merged: 19 of 20 authored. The 1 remaining is still open with a merge conflict. Same private Azure DevOps export, no public URLDate2026-08-12Sample sizen = 20

pull requests merged

Methodtables across two production databases documented by a dependency-free Python pipeline I wrote: it chunks schema for LLM analysis, then merges curated, AI-generated and rule-derived explanations into a searchable wiki, tagging each explanation with which of the three produced it. Figure as stated on my resume, no public URLDate2026-08-12

tables documented

Python
Claude API
MCP
Azure DevOps

Also shipped

Three more projects I’ve shipped, without a ledger. They’re smaller than the case studies above, and I haven’t sourced numbers for any of them, so this list says what each one is and nothing more.

local-qwen-coder-turboquant

Windows setup notes and scripts for running Qwen3-Coder-30B-A3B-Instruct as a local, coding-only, OpenAI-compatible backend on an 8 GB NVIDIA GPU (RTX 3060 Ti).

Built independently

PowerShell
Shell

Pixamart

A wallpaper application built with Flutter, using the Pexels API: categories, search, Firebase Auth including Google Sign-In, offline capability, and cloud sync.

Built independently

Dart
Flutter
Firebase
C++
CMake

repo (opens in new tab)

Chat-with-Code

A CLI that answers questions about any codebase using local Ollama models with LangChain retrieval-augmented generation, indexing a target folder by chunking and embedding source files into a local Chroma vector database.

Built independently

Python
Ollama
LangChain
Chroma

repo (opens in new tab)