AI Development Portfolio — USA

Challenge

BrightPath's support team was manually answering the same 15-20 admissions and billing questions on WhatsApp and email, often with same-day replies but never faster, which was costing them enrollments to competitors who responded in minutes.

Approach

We built a WhatsApp-native AI agent trained on their course catalog and policies, connected to their CRM so it could pull real enrollment status, and set escalation rules so anything outside its confidence handed off to a human instantly.

Results

• 82% of routine queries resolved without a human
• Average first response time dropped from 4 hours to under 1 minute
• Support team reassigned to high-value follow-ups, lifting enrollment conversion 19%

Challenge

Producing a narrated, presenter-style video from a source document (a brief, a deck, a script) normally means stitching together a scriptwriting pass, a voice generator, an avatar/talking-head tool, and a separate video editor — four tools, four exports, hours of manual assembly.

Approach

Built a single browser workspace where you drop in a brief plus optional character photo and voice sample, and the pipeline handles casting, blocking, narration (Coqui XTTS-v2) and talking-head rendering (SadTalker) end to end on a GPU backend, with a RAG layer (LanceDB + Sentence-Transformers) so the script can ground itself in an uploaded PDF/PPTX/Excel doc instead of hallucinating details.

Results

• One workspace replaces a 4-tool manual pipeline (script, voice, avatar, edit)
• Grounds narration in an uploaded source document instead of freeform generation
• Runs on a Google Colab GPU backend behind a standard Flask app — no dedicated GPU server to maintain

Challenge

Recording a walkthrough, sketching a live diagram, and running a multi-person review call normally means juggling a screen recorder, a separate diagramming tool, a whiteboard app, and video-conferencing software — none of which talk to each other.

Approach

Built the recording and live-collaboration half of the same workspace ReelForge lives in: screen/camera/mobile capture with a whiteboard overlay, an AI-prompted diagram generator (prompt-to-Mermaid, exports SVG/PNG), a freehand whiteboard with handwriting-to-text, and a WebRTC meeting room for up to 8 participants — including a QR-code flow so a phone can join as a second camera angle.

Results

• Screen, camera, mobile and mic capture combined with a live whiteboard overlay in one recording console
• AI diagram generator turns a plain-language prompt into an exportable Mermaid diagram
• Built-in WebRTC meeting room (up to 8 participants) removes the need for separate conferencing software

Challenge

The dispatch team was overwhelmed by email inquiries regarding freight status, leading to 12+ hour response times and dissatisfied enterprise clients.

Approach

Developed a custom autonomous email agent integrated directly into their ERP. The agent reads incoming freight queries, checks live telemetry, and drafts/sends accurate status updates.

Results

• Automated 88% of incoming status inquiries
• Response time reduced from 12 hours to under 3 minutes
• Client satisfaction scores hit all-time highs in 2026

Challenge

Acquisition costs on Google Ads were spiraling out of control due to aggressive US bidding wars, making the patient acquisition model unprofitable.

Approach

Pivoted to a heavily segmented Performance Max strategy with strict ROAS targets, augmented by first-party patient LTV data fed back into Google via Offline Conversions.

Results

• Patient Acquisition Cost (PAC) reduced by 38%
• Scaled campaign spend by 2x while maintaining target ROAS through Q2 2026
• Increased patient LTV by targeting higher-quality segments

Challenge

Acquisition costs on Google Ads were spiraling out of control due to aggressive US bidding wars, making the patient acquisition model unprofitable.

Approach

Pivoted to a heavily segmented Performance Max strategy with strict ROAS targets, augmented by first-party patient LTV data fed back into Google via Offline Conversions.

Results

• Patient Acquisition Cost (PAC) reduced by 38%
• Scaled campaign spend by 2x while maintaining target ROAS through Q2 2026
• Increased patient LTV by targeting higher-quality segments