Building AI that serves the grassroots · NLP · CSR Tech · LLM Evaluation · Indic Languages
I operate at the unique intersection of Low-Resource Language NLP / LLM Evaluation and Field-Level Community Development (CSR/NGOs). By day, I design data-driven village development projects for Tata Steel UISL through TRCSC in Jharkhand. By night, I build AI pipelines, Bengali NLP models, and automated tools that solve real operational bottlenecks for underserved communities.
| Project | Description | |
|---|---|---|
| 🇮🇳 | Kalki | India's Open Intelligence Platform — open-source early-warning signals from public data |
| 🛡️ | PromptShield | Defensive LLM structured extraction validation middleware |
| 🧠 | Obsidian LLM Wiki Kit | LLM-powered knowledge base toolkit — build and query interlinked wikis |
| 📊 | Bengali Annotation Quality | Bengali sentiment annotation calibration with Cohen's Kappa & disagreement analysis |
| 🤝 | SW परिवार | Indian Social Work Knowledge Base — laws, schemes, judgments (offline-first) |
| ⚡ | OpenCode Research Agents | Multi-agent research orchestration inside OpenCode |
Full project catalogue below.
Projects are auto-organized by domain from my active repositories.
| Project | Description | Language |
|---|---|---|
| 🧠 obsidian-llm-wiki-kit | A prompt-engineering starter kit for building a self-growing personal knowledge base in Obsidian, implementing Andrej Karpathy's LLM Wiki pattern. | Python |
| 🗣️ bengali-annotation-job-alert | Automated job monitor scraping LinkedIn, Remotive & RSS for Bengali AI data annotation and NLP roles, with Telegram alerts via GitHub Actions. | Python |
| 🤖 ai-checkkit | No description provided. | Python |
| 🤖 kp-astrology | KP astrology Python library: Krishnamurti Paddhati engine with nakshatra/sub-lord chains, Vimshottari dasha, Placidus cusps, birth-time rectification, and KP horary (1-249). Swiss Ephemeris powered. | Python |
| 🤖 skillcheck | Validate Agent Skills (SKILL.md) directories — catch missing fields, malformed frontmatter, and broken file references before your agent silently loses a skill. | Python |
| 🤖 mcphealth | Health-check MCP (Model Context Protocol) servers - confirm availability, supported tools and schema integrity from the CLI. | Python |
| 🤖 opencode-research-agents | A multi-agent research orchestration system running inside OpenCode, using epistemic parallelism to save synthesized findings to Obsidian. | Shell |
| 🎨 social-media-manager | SocialCraft AI: Zero-install client-side AI tool generating platform-optimized posts in 13 Indian regional languages using the Gemini API. | HTML |
| 🧪 bengali-annotation-quality | Annotation quality assessment pipeline calibrating Bengali sentiment datasets with Cohen's Kappa, disagreement analysis, and bias evaluation. | Python |
| 🤖 mcpconfig | Validate MCP client configuration files (.mcp.json, claude_desktop_config.json, opencode mcp: blocks) — catch the structural mistakes that silently break MCP servers. Zero dependencies. | Python |
| 🤖 wikicheck-test-create | scope test | Mixed |
| 🤖 mcpschema | Validate MCP (Model Context Protocol) tool JSON Schema definitions — catch missing fields, invalid types, and silent failures before your AI agent calls them. | Python |
| 🛡️ promptshield | Defensive LLM structured extraction validation middleware — guards against prompt injections and repairs truncated JSON responses. | Python |
| 🧠 markdownmind | Autonomous knowledge graph synthesizer for Markdown vaults — structures links, clusters topics, detects orphan files, and generates summaries. | Python |
| 📊 kappabench | Agreement and consensus benchmarking framework for LLM-as-a-judge classifiers — calculates Cohen's Kappa, Fleiss' Kappa, and Krippendorff's Alpha. | Python |
| Project | Description | Language |
|---|---|---|
| 🔀 splitcheck | Detect data leakage (shared samples) across ML dataset split files — flag rows that appear in both train and test before they inflate your evaluation metrics. | Python |
| 🏷️ annocheck | Validate YOLO-format CV datasets for corrupted labels, out-of-bounds boxes, missing images, and class mismatches — zero-config CLI. | Python |
| 🏷️ labcheck | Validate text annotation datasets (CSV/JSONL) for label conflicts, empty values, and quality issues — one command, zero config. | Python |
| 📊 coco-summary | Zero-dependency CLI to inspect COCO annotation JSON files — class distribution, bbox stats, and quality flags in one command. | Python |
| 📊 cocostat | COCO dataset statistics from your terminal — inspect annotation files: class distribution, bbox stats, annotation density, and more. | Python |
| Project | Description | Language |
|---|---|---|
| 📄 sci-calc-20260903 | ⚡ Production-grade scientific calculator with dual API, rich CLI/REPL, comprehensive error handling, and 100% test coverage. | Python |
| 🔲 fencecheck | Detect unclosed Markdown code fences (``` and ~~~) — catch broken code blocks that silently swallow the rest of your document. Zero dependencies. | Python |
| 📄 linkrot | Detect broken internal Markdown links — scan your docs for dead file references and missing section anchors. | Python |
| 📦 chunkviz | Visualize recursive character text splitting from your terminal — see chunk boundaries, overlap regions, and size distribution for RAG optimization. | Python |
| 📄 docstruct | Validate Markdown heading structure — detects skipped levels, missing H1s, and structural inconsistencies in documentation. | Python |
| 📄 wikicheck | Validate Obsidian-style [[wikilinks]] — catch broken targets, missing heading anchors, malformed links, and orphan notes in one zero-dependency CLI. | Python |
| Project | Description | Language |
|---|---|---|
| ⚡ tokensnap | Estimate LLM API costs from your terminal — pipe a prompt, get cost estimates across 12+ providers in one command. | Python |
| Project | Description | Language |
|---|---|---|
| ✅ frontcheck | YAML frontmatter linter for Markdown files. | Python |
| 🧬 globalcheck | Detect silent mutation of module-level state in Python — find functions that append to, update, or patch shared globals instead of returning values, the shared-state leak behind untestable AI code. Zero dependencies. | Python |
| 🧯 broadcheck | Detect bare, broad, and empty except handlers in Python — catch the silent error-swallowing (except Exception: pass) before it masks production crashes. Zero dependencies. | Python |
| ✅ todolint | Scan codebases for TODO/FIXME/HACK/XXX/BUG markers — CI-ready blocker detection with JSON output. Zero dependencies. | Python |
| 🔁 dupcheck | Detect duplicated code blocks across a codebase — catch copy-paste and AI-generated duplicates before they become maintenance debt. Zero dependencies. | Python |
| 🌪️ cccheck | Flag high-cyclomatic-complexity functions in Python codebases — catch God functions and untestable spaghetti before AI-generated code ships. Zero dependencies. | Python |
| Project | Description | Language |
|---|---|---|
| 🌾 ngo-grants-radar | Automated grant call monitoring for NGOs — tracks open funding opportunities from institutional donors and delivers Telegram alerts via GitHub Actions. | Python |
| 💼 ngo-jobs-radar | Automated NGO and social sector job alert pipeline — monitors 6 platforms and delivers real-time Telegram notifications via GitHub Actions. | Python |
| 📚 sw-parivar | Indian Social Work Knowledge Base — offline-first reference vault of laws, schemes, judgments, and UGC-NET frameworks for field practitioners. | HTML |
| 🌾 Rituparno-Majumdar.github.io | Rituparno Majumdar - AI Specialist, CSR Manager & Indologist. Portfolio and daily build log. | HTML |
| 🌾 cbo-capacity-assessment | CLI toolkit to assess CBO capacity across governance, finance, programmes, HR, advocacy, and infrastructure metrics. | Python |
| 🌾 csr-grant-tracker | Zero-dependency CLI for CSR grant compliance, milestone tracking, and Schedule VII reporting. | Python |
| 🌾 csr-impact-report | CLI tool for CSR grant impact calculation and compliance reporting — generate donor-ready metrics. | Python |
| 🌾 ngo-field-assistant | Offline-first CLI tool for NGO and CSR field workers — manage beneficiary data, validate entries, and generate reports without internet | Python |
| 🌾 bis-standards-indexer | CLI tool to index, search, and manage Bureau of Indian Standards (BIS) references — offline-first for compliance workflows | Python |
| 🎙️ voice-reporting-companion | Offline-first voice-to-JSON health reporting tool for ASHA and Anganwadi workers — auto-parses speech into NHM child health forms and stores in IndexedDB. | HTML |
| 📦 pds-audit-portal | Automated audit engine for India's Public Distribution System — detects duplicate ration cards, allocation shortfalls, and suspicious cancellation hotspots. | Python |
| ⚖️ bail-draft-builder | Interactive BNSS Section 479 under-trial bail eligibility calculator — computes detention threshold and generates formal, court-ready bail petitions. | HTML |
| 🌿 forest-rights-claim | Offline-first FRA 2006 land claim mapping tool for tribal communities — trace GPS boundaries on Leaflet and generate official Schedule I Form A PDFs. | HTML |
| 🇮🇳 kalki | Kalki: India's Open Intelligence Platform — open-source early-warning signals from public data to foresee climate, health, and economic crises. | Python |
| 📊 ngo-mis-toolkit | CLI toolkit for NGO reporting, logframe management & CSR compliance tracking in India. Generate donor-ready Word reports and validate LFA indicators. | Python |
| Project | Description | Language |
|---|---|---|
| 🛠️ herdr-4way-demo-20260903 | No description provided. | Mixed |
| 🛠️ Kriya | No description provided. | Mixed |
| 🛠️ tablecheck | Validate Markdown table structure - detect malformed tables, broken alignment and inconsistent columns. | Python |
| 🛠️ shabdabodha | Sanskrit NLP toolkit for transliteration, multi-dictionary lookup, and morphological word analysis. | Python |
| 🛠️ timeofbirth | KP (Krishnamurti Paddhati) birth time rectification — reverse-engineer the true time of birth from life events, right in your browser. Single HTML file, offline, private, global. | HTML |
Practical code patterns, guides, and configurations I reach for repeatedly — auto-posted alongside each new project.
| Date | Title | Description |
|---|---|---|
| 2026-08-29 | stubcheck Quick Guide | Detect silent no-op function stubs (pass/ellipsis bodies) that ship to production and return None — the unimplemented logic AI leaves behind in callbacks, handlers, and overrides |
| 2026-08-28 | assertcheck Quick Guide | Detect assert statements used for runtime validation — catch code that silently vanishes under python -O before it ships to production |
| 2026-08-20 | globalcheck Quick Guide | Detect silent mutation of module-level state in Python — catch functions that results.append(...) to shared globals instead of returning values, the shared-state leak behind untestable AI code |
| 2026-08-19 | broadcheck Quick Guide | Detect bare, broad, and empty except handlers — catch except Exception: pass silent error-swallowing before it masks production crashes |
| 2026-08-18 | mutablecheck Quick Guide | Detect mutable default arguments — catch def f(items=[]) shared-state aliasing before AI-generated code ships |
| 2026-08-16 | shadowcheck Quick Guide | Detect identifiers that shadow Python builtins — catch list = [] and def id() before runtime TypeErrors far from the binding site |
| 2026-08-15 | cccheck Quick Guide | Flag high-cyclomatic-complexity functions in Python — catch God functions and untestable spaghetti before AI-generated code ships |
| 2026-08-14 | dupcheck Quick Guide | Detect duplicated code blocks across a codebase — catch copy-paste and AI-generated duplicates before they become maintenance debt |
| 2026-08-13 | todolint Quick Guide | Scan codebases for TODO/FIXME/HACK/XXX/BUG markers — catch unfinished work before it ships, CI-ready blocker detection |
| 2026-08-12 | chunkviz Quick Guide | Visualize recursive character text splitting — see chunk boundaries, overlap regions, and size distribution before you tune your RAG pipeline |
| 2026-08-10 | mcpconfig Quick Guide | Validate MCP client configuration files — catch the structural mistakes that silently break your agent's tools |
| 2026-08-11 | fencecheck Quick Guide | Detect unclosed Markdown code fences — catch broken ``` and ~~~ blocks that silently swallow the rest of your document |
| 2026-08-09 | splitcheck Quick Guide | Detect data leakage across ML dataset split files — flag shared rows before they inflate evaluation metrics |
| 2026-08-08 | skillcheck Quick Guide | Validate Agent Skills (SKILL.md) directories — catch missing fields, broken frontmatter, and broken file references |
| 2026-08-03 | wikicheck Quick Guide | Validate Obsidian-style [[wikilinks]] — catch broken targets, missing heading anchors, and orphan notes |
| 2026-07-28 | linkrot Quick Guide | Detect broken internal Markdown links — find dead file references and missing section anchors |
| 2026-07-24 | labcheck Quick Guide | Validate text annotation datasets — catch label conflicts, empty values, and class imbalance |
| 2026-07-19 | Multi-Agent Research with OpenCode | Launch parallel research agents inside OpenCode — epistemic parallelism for robust synthesis |
| 2026-07-19 | annocheck Quick Guide | Validate YOLO CV datasets in one command — catch corrupted labels before training |
| 2026-07-20 | cocostat Quick Guide | COCO dataset statistics from your terminal — inspect annotation files instantly |
| 2026-07-21 | TokenSnap Quick Guide | Estimate LLM API costs from your terminal — pipe any prompt |
| 2026-07-22 | COCO Summary Quick Guide | Inspect COCO annotation datasets from terminal — class distribution and quality flags |
| 2026-07-23 | docstruct Quick Guide | Validate Markdown heading structure — catch skipped levels and missing H1s |
| 2026-07-23 | mcpschema Quick Guide | Validate MCP tool JSON Schema — catch missing fields before your agent calls them |
- ❗ Opened issue #4052 in herdrdev/herdr
- ❗ Opened issue #109971 in NousResearch/hermes-agent
- 🗣 Commented on #109727 in NousResearch/hermes-agent
- ❗ Opened issue #4028 in herdrdev/herdr
- 🚀 Published release v1.0.0 — Production-Grade Scientific Calculator & Rich CLI in Rituparno-Majumdar/sci-calc-20260903
I'm always excited to discuss low-resource language NLP, civic automation tools, biodiversity tracking, or practical LLM integrations for social impact.
- 🏠 Portfolio: rituparno-majumdar.github.io
- 💼 LinkedIn: linkedin.com/in/rituparno
- 📧 Email: majumdarrituparno@gmail.com
- 📍 Base: Jamshedpur, Jharkhand, India
- 🌐 BIS Standards: BIS Resource Person — Indian standards for AI & data
"Building scalable digital workflows so that physical field operations can impact more lives."