📍 Sunnyvale, CA | 🛠️ Building AI infrastructure | 🎓 IIT Bombay CS | 💼 Software Engineer @ LinkedIn

Ship fast, ship small, ship often — modular tools that compose into powerful systems.

📊 MarketOnePager (closed source) – Multi-agent market research platform.

• 🧠 pith – AI-native task management for humans & agents. MCP-first, CLI-native, self-hosted.
• ⚡ weav – In-memory context graph database written in Rust. Sub-10ms retrieval with vector search, temporal modeling, and token budgeting.
• 🤖 sweteam – Orchestrates your coding agents, so you don't have to.
• 🔁 ralph-py – Minimal CLI for orchestrating Ralph loops across coding agents.
• 🕷️ agent-crawl – High performance, lightweight and typesafe web crawler built for LLM agents.
• 📐 awesome-system-design – 30 production-grade system design documents for interview prep — from URL shorteners to AI inference platforms. Scale estimations, architecture diagrams, tradeoffs, and interview tips.

• 🧩 mcp-tool-router – Aggregates tools from multiple MCP servers into one
• 🧪 mcp-server-mock – Programmable mock MCP server for integration testing
• 🔍 mcp-schema-lint – CLI linter for MCP tool/resource schemas
• 🚦 mcp-rate-guard – Protocol-level rate limiter middleware for MCP
• 🏥 mcp-healthcheck – Programmatic health/liveness probe for MCP servers
• ⚙️ mcp-config-ts – CLI to discover, add, validate, and sync MCP server configs
• 📋 mcp-audit-log – Structured audit logger for MCP tool calls
• 📝 agent-scratchpad – Lightweight key-value scratchpad for agent reasoning
• 🔄 tool-call-retry – AI-specific retry wrapper with circuit breaker for tool calls
• 📐 tool-output-guard – Runtime validator for tool execution results against schemas
• 💰 tool-cost-estimator – Estimate token cost of tool definitions in the context window
• 📨 multimodal-msg – Provider-agnostic multimodal message builder

• 🔁 llm-retry – Smart retry orchestrator for LLM output parsing and validation
• 📼 llm-vcr – Record and replay LLM API calls for deterministic testing
• 🧹 llm-sanitize – Bidirectional I/O sanitizer middleware for LLMs
• 💾 llm-response-cache – Prompt-hash-keyed response cache with model-aware invalidation
• 📉 llm-regression – Semantic regression testing for prompt versions
• 🧽 llm-output-normalizer – Strip markdown fences and extract clean data from raw LLM output
• ✅ llm-eval-lite – One-function LLM evaluation with heuristic and model-graded assertions
• 💲 llm-cost-per-test – Jest/Vitest reporter for per-test-case LLM API costs
• 🔥 llm-chain-profiler – Flame-chart latency profiler for LLM chains
• 🐤 llm-canary – Invisible canary tokens for prompt leakage detection
• 📒 llm-audit-log – Tamper-evident compliance-ready audit logging for LLM I/O
• 🔀 llm-dedup – Coalesce semantically similar in-flight LLM requests
• 🧲 llm-semantic-cache – Self-hosted semantic cache using local embeddings
• 📄 llms-txt – Auto-generate and serve llms.txt for AI agent discoverability

• 🔬 prompt-lint – Static analysis for LLM prompts
• 🔀 prompt-diff – Semantic diff engine for prompts
• 🧹 prompt-dedup – Detect near-duplicate prompts via content hashing and similarity
• 📈 prompt-drift – Detect silent LLM output changes over time via semantic drift analysis
• 🚩 prompt-flags – AI-native feature flags for prompt variants and model selection
• 🧬 prompt-inherit – Composable inheritable prompt builder with TypeScript types
• ✂️ prompt-optimize – Compress prompts to use fewer tokens via deterministic heuristics
• 💵 prompt-price – Pre-flight cost estimation for any prompt and model combination
• 📸 prompt-snap – Jest-like snapshot testing for LLM outputs with fuzzy matching
• 🏷️ prompt-version – Local-first git-friendly prompt versioning with semver

• 🧱 chunk-smart – Structure-aware text chunker for RAG pipelines
• 📐 chunk-overlap-optimizer – Analyze chunk boundaries and recommend optimal overlap size
• 📦 context-packer – Optimally pack retrieved chunks into an LLM context window
• 💰 context-budget – Token budget allocator for LLM context windows
• 🪟 sliding-context – Provider-agnostic sliding window context manager for LLMs
• 🏗️ rag-prompt-builder – Compose RAG prompts from chunks with automatic metadata injection
• 📎 rag-cite – Extract and verify inline citations from LLM responses
• 📊 rag-eval-node-ts – Lightweight RAG evaluation metrics for CI/CD pipelines
• 🏅 rerank-lite – Lightweight retrieval reranker using cross-encoder scoring
• 🔗 fusion-rank – Reciprocal Rank Fusion for combining multiple retriever results
• 🔢 sparse-encode – Generate BM25 and TF-IDF sparse vectors in JavaScript
• 💎 embed-cache – Content-addressable embedding cache with deduplication and TTL
• 🔮 embed-cluster – Cluster embeddings into topics with automatic labeling
• 📉 embed-drift – Monitor embedding distribution shifts over time
• 📑 docling-node-ts – Convert documents to clean RAG-ready markdown in Node.js
• 📊 doc-table-extract – Extract tables from PDFs and images as structured JSON
• 📋 table-chunk – Extract and chunk tables preserving row/column structure

• 🕸️ kg-extract – Extract entity-relationship triples and build knowledge graphs
• 🔗 entity-resolve – Deduplicate and merge entity mentions across documents
• 🧠 memory-dedup – Semantic deduplication of agent memory entries
• 📝 md-to-data – Parse LLM markdown responses into typed JSON objects
• 📐 schema-from-text – Generate Zod schemas from natural language descriptions
• 🌉 schema-bridge – Write one Zod schema, get provider-specific structured output configs
• 🏷️ label-score – Inter-annotator agreement metrics in JavaScript
• 🧪 synthdata-gen – Generate and validate synthetic training data using any LLM
• 👤 synth-personas – Generate diverse synthetic user personas for AI testing
• 🎯 fewshot-gen – Generate diverse few-shot test cases from seed examples
• 📂 eval-dataset – Version-controlled eval dataset manager for LLM testing

• ⚡ ai-circuit-breaker – Circuit breaker for AI API spend management
• 💲 ai-cost-compare – Compare cost of running prompts across models and providers
• 🏦 ai-chargeback – Tag and allocate AI API costs by team, project, or feature
• 📈 ai-spend-forecast – Predict future AI API spending from historical usage
• 🔑 ai-keyring – Manage, rotate, and health-check AI API keys across providers
• 🏥 ai-provider-healthcheck – Monitor AI provider endpoint latency and availability
• 💰 model-price-registry – Auto-updating registry of LLM pricing across providers

• 🛡️ content-policy – Declarative business-rule content policy engine for LLMs
• 👻 hallucinate-check – Heuristic hallucination detection for Node.js
• 🚫 jailbreak-heuristic – Zero-dependency jailbreak attempt classifier using pattern matching
• 🧪 ai-output-assert – Rich assertion library for LLM outputs as Jest/Vitest matchers
• 📊 output-grade – Heuristic LLM output quality scoring without calling another LLM

• 🌊 stream-tokens – Aggregate streaming LLM tokens into semantic units
• ✅ stream-validate – Progressive Zod validation for streaming LLM responses
• 🎙️ voice-turn – Turn-taking manager for voice AI conversations
• 🔊 tts-queue – TTS audio streaming manager with sentence-boundary queuing
• 🎵 audio-chunker – Chunk audio streams into transcription-ready segments
• 🎬 vidsnap-ai – Smart video frame sampler for vision AI analysis
• 🖼️ vision-prep – Resize and optimize images for vision LLM APIs

• 🎨 ai-terminal-md – Terminal renderer optimized for AI markdown patterns
• ⏳ ai-spinner – AI-aware terminal progress indicators with token and cost display
• 🔀 ai-diff – Compare LLM responses across models with semantic diffs
• 📏 ai-rules-lint – Linter and validator for AI instruction files
• 📁 ai-file-router – Auto-detect file type and route through optimal parsing pipeline
• 🚀 ai-env-init – Bootstrap all AI config files from a single questionnaire
• 🔐 token-fence – Token budget enforcement middleware with intelligent truncation
• 🌳 convo-tree – Tree-structured conversation state manager for branching chats
• 🗜️ convo-compress – Incremental sliding-window chat compressor with rolling summaries
• 🧬 codebase-ctx – Generate AI-optimized codebase summaries via static analysis
• 📦 npm-package-skill – Claude Code skill that scaffolds production-grade npm packages

• AI developer infrastructure – Modular, composable TypeScript libraries for every layer of the AI application stack
• Agent orchestration – Tools to make multi-agent workflows reliable and observable
• Context engines – From Rust-powered graph databases to smart chunking and retrieval primitives
• Side projects – MarketOnePager