I’m Nawap. I turn messy workflows into AI products people actually use, from the data and the checks to the interface.
Money, private data and records people rely on. Three client problems where the work went well beyond the model.
01 / BusinessFrameLead AI engineer
Who controls the money?
Campaigns provisioned paused, launched by a person
I built website analysis, agent workflows and Google Ads provisioning around explicit human gates at billing, policy review and launch.
BusinessFrame · 0:28Silent walkthrough
Website analysis for a sample business. The approval path is explained in the case study.
02 / On-Prem Legal AIAI engineer
Where does our data go?
Models, documents and answers stayed on the box.
A legal AI workspace on hardware the customer owns. Local models, document collections, cited answers and an audit trail. Legal retrieval quality remained open work at handover.
On-Prem Legal AI · 1:33Silent walkthrough
Local workspace walkthrough. Names and documents are synthetic.
03 / Receipt OCRAI engineer
What if the model gets it wrong?
Nothing is saved until a person has seen the fields beside the image.
A delivery receipt becomes a reviewable form. The person corrects the fields, the database enforces the quantity rules, and the original photo stays with the record.
Receipt OCR · 1:01Silent walkthrough
Demo footage, from capture to export. Figures shown are not measured client results.
Two more products, and the challenge that started it.
AI engineer · 2026
Operational email becomes an auditable case, with policy evidence and approvals tied to the exact proposed action.
Synthetic demo data. Invoice arithmetic checked by code.
Associate AI developer · 2024 to 25
A plain language question becomes SQL, a table or a chart. Validation and rewrites happen before the query runs.
Built at Ujyalo Ventures. Generated code runs in an E2B sandbox.
Hackathon build · 2024
A speaking practice loop built in 48 hours with speech recognition, an LLM examiner and generated speech.
Hackathon prototype. Winner of AI Crusade 2024.
Retrieval, structured data, SQL validation, and clear paths back to the source.
Python · LlamaIndex · PydanticAI · SQLConnected tools, reviewable decisions, and explicit points for human judgment.
Agents · RAG · MCP · IntegrationsThe interface, streaming responses, deployment, and the details around the model.
TypeScript · Next.js · FastAPI · DockerBuilding agent workflows and automations for client businesses. Open for the next engagement.
Led an AI-managed Google Ads workspace, agent fleets included. Engagement wrapped Aug 2026 with a five-star review.
Built the text-to-BI microservice that shipped to production.
Good things start with a conversation.
Got something in mind?