AI systems that reach production
Working software with evals and monitoring, built by someone who ships these systems in his own companies. Not another proof of concept.
Finigami runs AI-first document extraction. Litmus Check is my open-source AI test-generation and QA platform. Geology builds tooling for AI search-visibility (GEO), the discipline of making content findable to LLMs rather than just traditional search engines. All shipped; none was a proof of concept.
That production experience is what I bring to client work. RAG pipelines, document processing agents, AI sales workflows, recruiting automation, AI coding agents: the six implementation guides on this site weren’t written from theory. They came from the work. If you’ve read them, you know what the build looks like.
Engagements start with a scoping call, move into a defined sprint, and end with working software your engineering team can maintain. I can embed with your engineers or run the build independently. The stack is Python-first, with LLM and vector infrastructure chosen to fit the problem.
Learn more about how I work on the Work with me page.
What you get
Working software, not a proof of concept
A delivered, tested system your engineering team can maintain. Evals in place so you know it holds up in production, not just in the demo.
Generative AI search-visibility
AI-powered content and tooling to make your organization findable to LLMs, the emerging discipline of GEO (Generative Engine Optimization), built on the same approach that underpins Geology.
Document and data extraction at scale
Structured extraction from unstructured documents: contracts, research reports, financial filings. The same architecture that runs Finigami.
Agent workflows that replace manual processes
Sales qualification, recruiting, content workflows, built as proper agents with memory, tool use, and error handling, not Python scripts that break on edge cases.
How it works
Scoping call
30 minutes to understand the use case, the data, and what done looks like. If the problem fits, I draft a technical scope and timeline before we start.
Design sprint
Architecture, data pipeline design, and a thin vertical slice that proves the core assumption. Fast feedback loop before full build.
Build and iterate
Working in defined sprints with deliverables your team can test. No months of silence followed by a reveal.
Handoff with evals
Documented system, evaluation suite, and a handoff session with your engineers so the system doesn't become a black box only I understand.
Proof, not promises
All client stories →AI Employee Assistance Chatbot
Designed and built a chatbot for a 40,000-employee organization to address questions about policies, asset tracking, and other internal tasks. Integrated with an updated knowledge base with an admin panel and the ticketing interface to create and track support tickets.
Leading financial institution in India
AI Vision for Fraud Prevention
Built a semantic image search interface to detect and present potential fraudulent gold loan applications to auditors for real-time fraud prevention.
Leading gold loan provider
Auto-scaling ML Inference
Designed and deployed a dynamically auto-scaling application for low-cost inference of ML jobs on geospatial data using GCP Cloud Run.
Public markets investor in MENA
Go deeper
- AI Coding Agents in 2026: A Practical Implementation Guide
- How to Build an AI Sales Agent in 2026: A Practical Implementation Guide
- Agentic Workflow Automation in 2026: A Practical Implementation Guide
- How to Build an AI Recruiting Agent in 2026: A Practical Implementation Guide
- How to Build a RAG System in 2026: A Practical Implementation Guide
- Intelligent Document Processing in 2026: A Practical Implementation Guide
- AI Agents for Business: A Working Field Guide
- AI Voice Agents for Customer Service
- Autonomous Agents: The Conversation You Need to Have With Your Board
- Autonomous AI Agents: What Executives Need to Know Before Approving One
- MCP and Your Multi-Model Strategy
- MCP and the End of Vendor Lock-In (Mostly)
Questions
What can you build? +
RAG systems and retrieval pipelines, document processing and extraction agents, AI sales qualification workflows, recruiting automation, AI coding agents, and generative AI tooling for content and search-visibility (GEO). The six implementation guides on this site cover these categories in depth: RAG, agentic workflows, AI coding agents, document processing, AI sales agents, and AI recruiting agents.
Do you build or just advise? +
I build. Finigami runs AI-first document extraction in production. Litmus Check is my open-source AI test-generation and QA platform. Geology runs AI search-visibility tooling. The guides on this site weren't written from theory. If you need someone who advises on what to build without building it, the AI strategy advisory service covers that.
What's your stack? +
Python-first. LLM layer depends on the use case: I work with Anthropic, OpenAI, and open-source models. Vector infrastructure via Pinecone, Weaviate, or pgvector depending on what fits. For agents: LangChain, LlamaIndex, or from-scratch depending on the complexity. No vendor allegiances.
How do projects start? +
A 30-minute scoping call, then a written scope covering the architecture, timeline, and what you get at the end. I don't start the build until we've agreed on what done looks like. It makes the rest of the engagement faster.
Can you work with our engineers? +
Yes. Some clients want me running the build independently; others want me embedded with their team. Both work. Embedded engagements tend to leave the team with more capability at the end, since they've seen the architecture decisions get made in real time.
Let's talk
30 minutes, no slides. We'll work the specific decision you're facing.