Ask me anything

Problem

Manual tech-stack research across many companies is slow, inconsistent, and difficult to scale for outbound qualification and competitive analysis.

Solution

The workflow reads company names from a spreadsheet in batches, processes each company individually, gathers technology signals, and uses an OpenRouter-powered LLM to infer and structure each stack before writing results back to the sheet.

Architecture

Manual trigger starts a spreadsheet-driven pipeline. A loop processes up to six companies per run, calls an OpenRouter chat model for stack analysis, formats parsed output into structured fields, and appends results to the original spreadsheet for continuous enrichment.

Problem

Static documents are hard to query quickly. Teams spend time manually searching contracts, reports, and manuals instead of getting direct answers grounded in document content.

Solution

The system runs in two phases: indexing and runtime querying. On file add/update, it ingests documents, chunks text, creates embeddings, and stores vectors. At query time, it embeds the question, retrieves relevant chunks, and generates context-grounded responses.

Architecture

Event-driven indexing starts from Google Drive updates, followed by chunking (Recursive Character Text Splitter), OpenAI embeddings, and Pinecone vector storage. Runtime chat embeds user questions, retrieves top-matching chunks from Pinecone, and sends question + context to an OpenRouter chat model for final answers.

Problem

Lead research is repetitive and fragmented across search, websites, and social channels. Manual workflows are slow and often produce inconsistent or low-quality contact data.

Solution

The system automates company discovery, multi-source scraping, and contact verification. It gathers emails and social profiles, validates deliverability, and enriches each lead into a structured profile ready for outbound workflows.

Architecture

Three-phase pipeline: company discovery from provided names/domains, scraping agents for emails and social profiles, then a verification/enrichment stage that consolidates data into reliable records. The modular flow separates search, extraction, and validation for scalable operation.

Problem

High-volume inboxes create bottlenecks where important messages are delayed, triage is inconsistent, and manual sorting consumes significant time.

Solution

The workflow runs extraction and validation first, then classifies each message into AI, Books, Client, or Miscellaneous. It auto-labels all categories and sends tailored replies for AI, Books, and Client while keeping Miscellaneous as label-only.

Architecture

Triggered by new Gmail events, phase one uses an OpenRouter model for information extraction plus conditional logic to handle missing sender data and normalize fields. Phase two uses a second OpenRouter classification step with rule-based routing to Gmail labeling and automated response actions.

Problem

Manual X/Twitter research is slow and inconsistent, making it hard for content teams to reliably identify viral patterns across topics and time windows.

Solution

The pipeline runs on schedule or manual trigger, searches tweets by configured keywords, processes multiple pages, extracts engagement/content metadata, and ranks results to keep only the highest-performing posts.

Architecture

Three-phase workflow: setup/scheduling with search-term configuration and pagination, scraping loop with Twitter API retrieval and per-item extraction, then processing/storage with item splitting, engagement-based sorting, and sheet appends for downstream content operations.