Bright·Nwokoro
01AI Engineer · Lagos → Remote

Bright Nwokoro

Production AI assistants, shipped end-to-end.

I build RAG systems, embeddable chat widgets, and AI copilots for SaaS and startups. Full-stack. No hand-offs.

See live demosStart a project
Bright Nwokoro — Full-Stack + AI Engineer
Based in
Lagos → Remote
02About

Full-stack engineer. AI focus. Shipping bias.

I'm Bright, a full-stack engineer based in Lagos who builds AI systems that work in production — not just demos. My current focus is RAG pipelines and AI assistants: the kind that live inside your product, answer real user questions against real data, and don't hallucinate their way out of a job.

I've shipped across Node.js, Python, React, and Next.js, with working knowledge of the modern AI stack — vector databases (pgvector, Pinecone), LLM APIs (OpenAI, Claude, Gemini), orchestration (LangChain, LangGraph), embeddings, chunking strategies, evaluation, and the messy real-world parts: latency, cost, retrieval quality, and guardrails.

I work directly with founders. No agencies, no layers. You tell me the problem, I ship the thing.

01

AI Assistants

Embeddable chat widgets and copilots that understand your product and docs.

02

RAG Systems

Retrieval pipelines over PDFs, Notion, databases, and knowledge bases — with citations.

03

Full-Stack Delivery

Frontend, backend, deployment, monitoring. One person, start to finish.

04

Production Mindset

Evaluation harnesses, cost monitoring, latency budgets. Built to last past launch.

03Introduction

A word in person.

Twenty focused minutes. I'll hear your problem, give an honest read on approach and timeline, and you'll know whether we should work together.

·20 min · Free · No pitch

Book an intro call.

  • 01Your AI problem + constraints
  • 02Rough technical approach
  • 03Realistic timeline + cost

Pick a time on the next page — instant confirmation, Google Meet link generated automatically.

Opens in-page
04Engagements

Selected engagements.

Live AI systems I've built end-to-end — problem, approach, outcome. New engagements ship every few weeks.

D
Shipped 2026

RAG over PDFs with Citations

DocSage
Self-initiated · Live

Founders with large document sets need AI that cites its sources. Built a semantic-chunked RAG pipeline with hybrid retrieval (vector + keyword) and strict citation enforcement via structured output.

Key Achievements

  • •Per-query cost held under $0.002 via GPT-4o-mini
  • •pgvector over Pinecone to keep infra near zero for SMB corpora
  • •Inline citations with page numbers and confidence scores
Next.jsFastAPIpgvectorOpenAIVercelRailway
E
Shipped 2026

Drop-in AI Widget for Any Site

EmbedChat
Productized · Live

Shipped a ~30kb vanilla TypeScript widget with Shadow DOM isolation so it never collides with host CSS. Backend proxies to the LLM with configurable system prompts and optional RAG grounding.

Key Achievements

  • •One-line <script> install, configured via data-attributes
  • •Shadow DOM isolation — zero CSS collision with host sites
  • •Streaming responses from an edge-deployed proxy
TypeScriptShadow DOMHonoCloudflare WorkersClaudepgvector
J
Active product · v2 in 2026

AI-Powered Job Platform for African Tech

JobDork
Self-initiated · Live

Evolved a live job platform with AI-powered matching (embeddings over CVs + JDs), tailored cover-letter drafts, and LLM-summarized daily digests of top matches.

Key Achievements

  • •Semantic CV↔JD matching via pgvector similarity
  • •Per-user cover-letter drafting grounded in their CV + the JD
  • •Daily digest automation summarizing top matches
Next.jsTypeScriptPostgreSQLpgvectorOpenAIDockerAWS
05Selected Work

Featured projects.

Five AI systems shipped end-to-end. Each one solves a real problem I've heard founders ask for in the last ninety days.

01Case Study

DocSage — RAG over PDFs with Citations

A RAG chatbot over any PDF set, answering with inline source citations and confidence scores.

DocSage — RAG over PDFs with Citations
·The Challenge

Founders with large document sets need AI that cites its sources — not hallucinates them. Legal, policy, manuals, and API docs all require verifiable answers.

·The Approach

Built a RAG pipeline with semantic chunking, hybrid retrieval (vector + keyword), and strict citation enforcement via structured output. Chose Postgres + pgvector over Pinecone to keep infra near zero for small-to-medium corpora. GPT-4o-mini keeps per-query cost under $0.002.

·The Outcome

Live demo preloaded with Next.js, Stripe, and Kubernetes documentation. Visitors can ask technical questions and see inline citations to the exact PDF + page.

·Key Metrics
$0.002

per query

3-5s end-to-end latency

100%

citation coverage

Zero infra lock-in

Next.jsFastAPIpgvectorOpenAIRAGVercelRailway
02Case Study

EmbedChat — Drop-in AI Widget for Any Website

A single-script AI chat bubble that any SaaS can paste in and configure via data-attributes.

EmbedChat — Drop-in AI Widget for Any Website
·The Challenge

SaaS founders want AI chat on their marketing site or in their product — but don't want to integrate a full chat platform or fight CSS collisions.

·The Approach

Shipped a ~30kb vanilla TypeScript widget wrapped in Shadow DOM so it never conflicts with host-site CSS. Backend proxies to the LLM with per-site system prompts and optional RAG grounding over sitemap/Notion/docs URLs. Deployed on Cloudflare Workers for edge latency.

·The Outcome

Productized as a $1.5k–$5k offer. Demo deployed to a fake SaaS site where visitors can interact with the widget live — the widget is trained on the demo site's docs so the experience is believable in 10 seconds.

·Key Metrics

~30kb bundled

<100ms time-to-first-token

1-line install

SOC-2 friendly proxy

TypeScriptShadow DOMHonoCloudflare WorkersClaudepgvector
03Case Study

MeetingScribe — AI Meeting Summaries with Action Items

Uploads audio or transcripts and returns a structured summary with decisions, action items, and assigned owners.

MeetingScribe — AI Meeting Summaries with Action Items
·The Challenge

Teams want AI meeting notes but don't want to hand private conversations to SaaS platforms charging $20/user/month.

·The Approach

Whisper for transcription, then Claude with JSON-schema structured output to guarantee valid action-item extraction with owner + deadline fields. Structured output is the senior move — zero downstream parsing fragility.

·The Outcome

Self-hostable alternative to Otter/Fireflies. Summaries return in parseable JSON ready to be piped into email, Notion, or Linear.

·Key Metrics

JSON schema-guaranteed

$0.10–0.40 per meeting

Whisper + Claude pipeline

Supabase end-to-end

Next.jsWhisperClaudeSupabaseStructured Output
04Case Study

InboxTriage — AI Email Classifier and Drafter

Connects to Gmail, classifies unread emails, and drafts replies for the ones that need them.

InboxTriage — AI Email Classifier and Drafter
·The Challenge

Inbox overload is universal and nobody wants to give a SaaS full access to their Gmail. Teams want triage without giving up control.

·The Approach

Gmail OAuth with read-only scope. Claude Haiku for cheap, fast classification into urgent / needs-reply / FYI / newsletter / spam. Agent pattern drafts replies only for the 'needs reply' bucket and never auto-sends — the human stays in the loop.

·The Outcome

Self-hostable. Dashboard shows triaged inbox with one-click draft preview. Demo recorded on a burner Gmail so the public demo doesn't require live OAuth from visitors.

·Key Metrics
$0.001

per email classified

OAuth read-only scope

Agent-pattern drafting

Self-hostable

Next.jsGmail APIClaude HaikuOAuth 2.0Supabase
05Case Study

JobDork v2 — AI-Powered Job Platform

A live job platform for African tech professionals, evolved in 2026 with AI-powered matching, cover-letter drafts, and daily digests.

Live demo
JobDork v2 — AI-Powered Job Platform
·The Challenge

Generic job boards don't understand either the candidate's CV or the African tech market nuances. Matches are noisy; applications go unread.

·The Approach

Embed both the CV and every job description using OpenAI embeddings, store in pgvector, rank by cosine similarity. Tailored cover-letter drafts are grounded in both the CV and JD. A daily LLM-summarized digest surfaces the top 5 matches per user.

·The Outcome

Live at job-dork.brightnwokoro.dev — a production platform used by African tech professionals. The AI layer turns a basic aggregator into a personalized job concierge: rankings, drafts, and digests all grounded in real embeddings.

·Key Metrics

Embedding-based matching

Daily digest automation

Real users, real product

Lagos-built, global-ready

Next.jsTypeScriptPostgreSQLpgvectorOpenAIDockerAWS
06Skills

Tools of the craft.

AI first. Full-stack second. The stack I reach for when clients need something shipped end-to-end, not handed off between three freelancers.

AI & LLM
  • 01LLM Integration
  • 02RAG Pipelines
  • 03Agent Systems
  • 04Prompt Engineering
  • 05Evaluations
  • 06LangChain / LangGraph
Backend
  • 01Node.js (Express, NestJS)
  • 02Python (FastAPI, Django)
  • 03REST & GraphQL
  • 04WebSockets
  • 05Background Jobs
  • 06API Design
Frontend
  • 01React
  • 02Next.js
  • 03TypeScript
  • 04Tailwind CSS
  • 05Framer Motion
  • 06Shadow DOM Widgets
Data
  • 01PostgreSQL
  • 02pgvector
  • 03MongoDB
  • 04Redis
  • 05Pinecone / Qdrant
  • 06Embeddings
Infrastructure
  • 01AWS
  • 02Docker
  • 03Kubernetes
  • 04Terraform
  • 05CI/CD Pipelines
  • 06Cloudflare Workers
08Contact

Have an AI problem that needs shipping?

Currently taking on 2 new projects for Q2 2026. Typical engagement: 2–6 weeks, fixed-price, shipped end-to-end. I respond to every email within 24 hours.

·Write to mehello@brightnwokoro.dev
GitHubLinkedInTwitter
Taking 2 new projects · Q2 2026