I’m an AI engineer based in Lagos.

I’m Udochukwu Echefu, based in Lagos. I build Python backends and interfaces for support triage, contract analysis, and finance operations.

View projects Résumé

Projects

Kora Support Triage

A support queue that classifies WhatsApp and email cases, applies policy checks, and records the reviewer’s decision.

Python · FastAPI · React · Groq · SQLite

Open demo View source

Technical notes

Structured outputs route cases to six specialist teams. Deterministic guardrails, confidence thresholds, SLA controls, and SQLite memory support review. Model and human decisions are recorded in an audit log.

The project record includes a 100-case synthetic policy dataset, 132 backend tests and 11/11 Groq smoke cases passing. These are recorded results, not a live test status.

Kora support queue with conversation list and case review panel — Public demo with synthetic support cases.

Lenslayer

Built an evidence-led AI workspace that turns contract analysis into ongoing operational work. Teams can ask grounded questions across a contract portfolio, inspect the cited agreements and excerpts behind every answer, reconcile identity evidence, and move findings into human-owned inbox, task, calendar, verification, and reporting workflows.

Next.js · TypeScript · Cloudflare

Open workspace View source

WHT recovery overview with a case queue and evidence coverage — Recovery overview, shown with synthetic records.

WHT Recovery Control

Built an AI-assisted control workspace for Nigerian withholding-tax receivable recovery. DeepSeek extracts receipt fields through a strict schema with confidence scores, exact source quotes, and page provenance; deterministic rules reconcile evidence to ledger cases, while reviewers retain control over recognition, disputes, and closure. Cloudflare D1, R2, and audit events keep the workflow durable and inspectable.

DeepSeek · TypeScript · Cloudflare D1 + R2

Open workspace View source

ContractGuard

Built a document-intelligence workspace that ingests PDF, DOCX, TXT, and scanned documents, retrieves evidence with ChromaDB and all-MiniLM-L6-v2 embeddings, and uses Groq through LangChain to generate citation-linked risks, obligations, negotiation questions, and grounded follow-up answers.

Streamlit · LangChain · ChromaDB

Open app View source

Production RAG Pipeline

Built a production-minded RAG service with FastAPI and PostgreSQL. It ingests and versions TXT, Markdown, and PDF sources, then combines ParadeDB BM25 with pgvector semantic search using Reciprocal Rank Fusion and page-level provenance. Recorded SciFact results: 84.44% Recall@10, 66.83% MRR@10, and 70.75% nDCG@10 across 5,183 documents and 10,980 chunks.

Python · FastAPI · PostgreSQL · pgvector · ParadeDB

View source

About

I work across APIs, retrieval pipelines, and the interfaces people use to review model output.

My projects have grown from document question-answering into support and finance workflows, with validation, persistent records, and human review built into the application.

Engineering profile

FastAPI application systems
Structured outputs and retrieval
Guardrails, tests, and audit trails
Operational React interfaces

Download résumé

I turn customer and document workflows into auditable, human-in-the-loop software using Python, FastAPI, retrieval, structured LLM outputs, and product design.

AI engineering stack

Models + orchestration: Claude, ChatGPT, Kimi, DeepSeek, Groq-hosted LLMs, structured outputs, prompt design, LangChain, and OpenAI-compatible APIs.
Retrieval + grounding: RAG, ChromaDB, Hugging Face embeddings, all-MiniLM-L6-v2, evidence retrieval, and citation pipelines.
Reliability + evaluation: Deterministic guardrails, synthetic evaluations, confidence thresholds, schema validation, human review, and model and decision audit trails.
Application + data: Python, FastAPI, SQLAlchemy, Alembic, PostgreSQL, SQLite, REST APIs, background workers, and durable workflow records.
Interfaces + deployment: Next.js, React, TypeScript, Streamlit, Vite, Tailwind CSS, Cloudflare Workers and R2, and Railway.

Selected engineering milestones

Grounded retrieval

Implemented LangChain prompt pipelines, document chunking, Hugging Face embeddings, and Chroma similarity search for evidence-grounded question answering.

Deployed retrieval with conversational memory

Deployed a Streamlit RAG application with cached vector indexes, PDF ingestion, session history, and query rewriting for ambiguous follow-up questions.

Auditable AI operations

Shipped Kora with structured model outputs, deterministic guardrails, SQLite customer memory, persisted audit trails, operational controls, automated tests, and Railway deployment.

From AI analysis to an operational workspace

Expanded document intelligence into Lenslayer: portfolio-wide grounded Q&A, citation-backed findings, identity-evidence review, and human-owned decision workflows across inbox, tasks, calendar, verification, and reports.

Contact

I’m open to backend, AI application, and product-engineering opportunities where I can build reliable Python systems and ship customer-facing software.

udochukwuechefu10@gmail.comGitHub ↗Résumé ↓