Evidence Signal
AI chat · JD analyzer · project evidence routing
Project Deep Dive
Latif AI Guide
This website is the proof: a functional AI platform with evidence-backed chat, project routing, and real-time job-fit analysis, built to demonstrate what production AI systems look like
The interactive layer of this personal site: evidence-backed chat, live job-fit analysis, and project navigation.
Interactive experience designAI agent designPlatform product
The Problem
What problem was worth solving?
I build things to solve a specific pain, not to decorate a portfolio. This is the pressure that made this project necessary.
A static resume cannot show how someone thinks, what they have built, or their approach to solving problems.
Evaluators and hiring teams want to interact with evidence — ask questions, see real systems in action, and get personalized analysis.
Latif AI Guide was built to solve this: an interactive experience that lets people explore, ask, and evaluate in real time.
The Build
What I built to solve it
A full-stack AI platform that powers this website. Latif AI Guide composes intelligent responses using an 8-layer system that adapts to what the visitor is asking about — with different reasoning approaches depending on whether someone is exploring, evaluating fit, asking for evidence, or planning next steps. The job description analyzer runs a 3-phase pipeline: extract requirements, find matching evidence, then evaluate fit with confidence grading. Responses stream in real time so the site feels inspectable, not static.
Adaptive AI system
8-layer intelligence with 4 reasoning modes
The AI composes its response differently depending on what the visitor needs — exploring broadly, evaluating fit, requesting evidence, or planning next steps. Adapts its depth based on whether it is in text chat or video mode.
Job-fit analysis
3-phase pipeline: extract → match → evaluate
Extracts requirements from job descriptions, finds matching evidence from real project experience, then evaluates fit with confidence grading (strong match, partial match, or gap) for each requirement.
Safety and quality
5-layer output filtering
Every AI response passes through 5 quality and safety checks before reaching the user — preventing prompt leaks, low-quality outputs, and persona inconsistencies.
Performance
Real-time streaming with persistent memory
Responses stream in real time so the visitor never waits. Conversation history is saved in the background without slowing down the AI. The system degrades gracefully if the database is temporarily unavailable.
Value
What changed because it existed
This is the clearest evidence of practical value, system leverage, and execution quality.
AI chat
Evidence-backed conversational AI
Latif AI Guide answers questions using real evidence from a knowledge graph — not generic AI responses. Per-visitor rate limiting and safety filtering on every response.
JD analysis
Real-time job-fit scoring
Paste a job description and get an instant analysis showing how experience maps to each requirement — with confidence ratings and evidence for each match.
Video avatar
Live AI video conversation
The video layer uses the same concise, evidence-backed public-site guide posture. Available to authenticated users.
The platform is the proof
This website demonstrates the work
Every feature on this site — the AI chat, the job analyzer, the video avatar, the streaming architecture, the security design — is a working demonstration of production AI engineering.
Tech Stack
What it runs on
The stack matters here because it reflects design choices, constraints, and how the system was intended to scale or integrate.
Next.js 16 + React 19 + TypeScript 5.9FastAPI on Azure App ServiceAzure OpenAI GPT-4.1text-embedding-3-large (dimensions=1024, pinned)pgvector on PostgreSQLServer-Sent Events (SSE streaming)JWT + Azure Key VaultTavus CVI (video avatar)Tailwind CSS 4
Build Story
How the thinking unfolded
This is the reasoning path behind the output, not just the finished artifact.
01
The platform is the proof
This website was not built as decoration. It was built to demonstrate production AI engineering — adaptive conversation, real-time streaming, video integration, evidence-backed responses, and security design — all running live.
02
Adaptive AI that understands context
The chat system adapts to what each visitor needs. It recognizes whether someone is exploring, evaluating fit, requesting evidence, or planning next steps — and adjusts its reasoning and depth accordingly. Video mode keeps responses conversational; text chat goes deep.
03
Built for real-world performance
Responses stream in real time. Conversation history is saved in the background without slowing down the AI. The job analyzer manages evidence budgets to keep analysis predictable. The system degrades gracefully if any component is temporarily slow.
04
Security built in from day one
Five quality and safety checks run before any AI response reaches a visitor. The public chat is rate-limited. Video sessions require authentication. All credentials are managed through Azure Key Vault — never in code.
Capability Signal
What this project demonstrates
Each project is a proof point. These are the capabilities it most clearly reveals.
AI System Design
Designed an adaptive 8-layer AI system with 4 reasoning modes, context-aware responses, and channel-specific formatting — all composed dynamically for each conversation.
Real-Time Streaming
Built a streaming pipeline for job-fit analysis with parallel evidence gathering, budget management, and background data persistence — responses start appearing immediately.
AI Safety and Quality
Implemented 5 sequential output checks (safety filtering, quality gates, persona consistency) before any AI response reaches the visitor.
Evidence-Based AI
Knowledge graph with semantic search powers evidence-backed responses. Conversation memory persists across sessions with graceful timeout handling.
Secure Platform Design
Authentication-gated video, per-visitor rate limiting, centralized credential management, and no internal system details exposed to users.
Continue
Want to go deeper?
Ask the Latif AI Guide about the architecture, the commercial logic, or what this project says about how I approach hard problems.