SSohaima R.

Team Lead · CodeXcape

Sohaima R. — AI & full-stack team lead who ships.

Team lead at CodeXcape, working across the whole line — the model, the service around it, and the screen a person actually touches. She has led delivery for medical, fintech, AI calling, real estate, health & care and e-commerce products: AI systems that have to be right often enough to trust, inside products that hold up once real users arrive.

Hire me on FiverrEmail me
  • 6 industries delivered
  • AI agents, voice & RAG
  • Team lead, end to end
Sohaima R., AI & Full-Stack Team Lead at CodeXcape

Currently

Leading AI builds at CodeXcape

Shipping AI used to take a research team and half a year.
Now it takes one engineer who finishes.

The team lead behind CodeXcape’s AI builds.

At CodeXcape, Sohaima leads delivery from first call to handover: she scopes the work, sets the architecture, runs the sprints, reviews the code, and is the person a client talks to when something has to be decided. The team builds; she makes sure what it builds is the thing that was needed.

A demo answers a question. A product answers it on a Tuesday afternoon, for a user who is in a hurry, with the data that was loaded last night.

  • Scopes the work with the client
  • Sets the architecture and AI approach
  • Runs sprints and reviews every PR
  • Owns launch and handover
MedicalCase study →FintechCase study →AI Calling AgentCase study →Real EstateCase study →Health & CareCase study →E-commerceCase study →

How a project moves under her lead.

The same six stages on every engagement, so a client always knows where the work is and what comes next.

  1. 01

    Discover

    Workshops with the people who feel the problem

  2. 02

    Architect

    Data model, AI approach, risks named early

  3. 03

    Build in sprints

    Two-week slices, demoed every time

  4. 04

    Test & harden

    Evals for AI, tests for everything else

  5. 05

    Launch

    Deployed, monitored, rolled out in stages

  6. 06

    Hand over

    Documented so your team can own it

Six industries. Six systems she led into production.

The problem the client had, the system her team built — drawn as a flowchart — and what it does now. Client names are left out until a client says otherwise.

AI clinical documentation workflow for a medical practice

Medical

Clinical documentation that writes itself — and gets checked.

Fintech KYC onboarding and payment risk dashboard

Fintech

Onboarding and risk checks without the manual queue.

AI calling agent handling inbound customer calls

AI Calling Agent

A voice agent that answers, qualifies and books — 24/7.

Real estate property listing and lead management platform

Real Estate

Listings, leads and viewings in one platform.

Home care and telehealth coordination platform

Health & Care

Care plans, visits and family updates — on one app.

E-commerce storefront with AI shopping assistant

E-commerce

A store that answers questions and recommends what fits.

Flowchart keyInputSystemAI stepHuman checkOutcome
01Medical

AI clinical notes for a multi-specialty practice

The problem

Doctors were spending the end of every day typing up notes from memory. Notes were late, inconsistent between specialties, and coding errors were found at billing — weeks after the visit.

Her role as team lead

Led a team of four: ran discovery with the clinicians, designed the redaction step and review screen, and owned the HIPAA-minded data flow.

AI clinical documentation workflow for a medical practice

How the system works

  1. InputConsultationAudio or typed notes
  2. SystemTranscribe & redactPHI masked before the model
  3. AI stepDraft noteLLM fills the specialty template
  4. Human checkDoctor reviewApprove or edit in one screen
  5. OutcomeEHR + codingSigned note and suggested codes

What it delivers

  • Notes drafted by the end of the visit instead of the end of the day
  • One template per specialty, so notes read the same across doctors
  • Nothing reaches the record without a doctor signing it

Where the build effort went (estimate)

  • AI / ML 35%
  • Backend 30%
  • Frontend 20%
  • Integrations 15%
  • Python
  • Whisper
  • OpenAI API
  • FastAPI
  • React
  • PostgreSQL
02Fintech

KYC onboarding and transaction risk engine

The problem

Every new account sat in a manual review queue. Document checks were done by eye, risk rules lived in a spreadsheet, and good customers waited days while risky ones slipped through on busy afternoons.

Her role as team lead

Led architecture and a team of five; worked directly with the compliance lead to turn policy into rules the system could enforce and explain.

Fintech KYC onboarding and payment risk dashboard

How the system works

  1. InputSign-upID, selfie, business docs
  2. AI stepDocument AIExtract and match fields
  3. SystemRisk scoringRules + model, explained
  4. Human checkAnalyst queueOnly the edge cases
  5. OutcomeAccount liveAudit trail kept

What it delivers

  • Clear cases approved automatically, edge cases routed to an analyst
  • Every decision stored with the reason it was made, for the auditor
  • Risk rules editable by the compliance team without a deploy

Where the build effort went (estimate)

  • AI / ML 25%
  • Backend 35%
  • Frontend 20%
  • Integrations 20%
  • TypeScript
  • Node.js
  • Python
  • PostgreSQL
  • Stripe
  • React
03AI Calling Agent

AI voice agent for inbound and outbound calls

The problem

Calls after hours went to voicemail, and most were never returned. The sales team spent mornings calling back leads who had already gone to a competitor.

Her role as team lead

Led the build end to end: telephony, the agent’s tool design and guardrails, latency tuning, and the call dashboard the client’s team uses daily.

AI calling agent handling inbound customer calls

How the system works

  1. InputPhone callInbound or scheduled outbound
  2. SystemSpeech-to-textStreaming, low latency
  3. AI stepVoice agentLLM with narrow, typed tools
  4. SystemBook / CRMCalendar slot, lead record
  5. Human checkHandoffWarm transfer to a human

What it delivers

  • Every call answered, any hour, in a natural voice
  • Leads qualified and booked straight into the calendar and CRM
  • Full transcript and summary on every call for the team to review

Where the build effort went (estimate)

  • AI / ML 40%
  • Backend 25%
  • Frontend 15%
  • Integrations 20%
  • Twilio
  • Deepgram
  • OpenAI Realtime
  • Node.js
  • MongoDB
  • React
04Real Estate

Property platform with AI lead matching

The problem

Agents were juggling listing portals, WhatsApp threads and a spreadsheet of buyers. Enquiries were answered late, and nobody could say which listing a lead had actually asked about.

Her role as team lead

Led a team of four through discovery with agents on the floor, the data model for listings and leads, and the matching logic.

Real estate property listing and lead management platform

How the system works

  1. InputEnquiryPortal, site, WhatsApp
  2. SystemLead captureOne record per buyer
  3. AI stepAI matchingBuyer needs ↔ listings
  4. Human checkAgent follow-upViewing booked
  5. OutcomeDeal pipelineOffer to close

What it delivers

  • All enquiries land in one pipeline with the listing attached
  • Buyers matched to new listings the day they go live
  • Managers see the pipeline per agent without asking for a report

Where the build effort went (estimate)

  • AI / ML 20%
  • Backend 30%
  • Frontend 30%
  • Integrations 20%
  • Next.js
  • TypeScript
  • Node.js
  • PostgreSQL
  • OpenAI API
  • WhatsApp API
05Health & Care

Home-care coordination and telehealth platform

The problem

Carers were scheduled on paper, visit notes came back in photos of notebooks, and families phoned the office to ask how a parent was doing.

Her role as team lead

Led product and engineering for a team of five, including the offline-first carer app and the role model for carers, coordinators and families.

Home care and telehealth coordination platform

How the system works

  1. InputCare planNeeds, meds, schedule
  2. SystemSchedulingCarer matched to visit
  3. Human checkVisit appCheck-in, tasks, notes
  4. AI stepAI summaryFlags changes in condition
  5. OutcomeFamily portalUpdates and video calls

What it delivers

  • Visits scheduled, checked in and recorded on the carer’s phone
  • Changes in a client’s condition flagged to the coordinator the same day
  • Families see updates without calling the office

Where the build effort went (estimate)

  • AI / ML 15%
  • Backend 30%
  • Frontend 35%
  • Integrations 20%
  • React Native
  • React
  • Node.js
  • PostgreSQL
  • WebRTC
  • OpenAI API
06E-commerce

AI shopping assistant and storefront

The problem

Shoppers asked the same sizing and delivery questions on chat all day, product search missed anything not typed exactly, and support was the bottleneck at every sale.

Her role as team lead

Led a team of four across storefront, catalogue pipeline and the assistant; owned performance on mobile, where most of the traffic was.

E-commerce storefront with AI shopping assistant

How the system works

  1. InputShopperSearch or chat
  2. SystemSemantic searchCatalogue embedded
  3. AI stepAI assistantSizing, stock, orders
  4. SystemCheckoutCart and payment
  5. OutcomeOrder + supportTracking, returns

What it delivers

  • Search that understands what a shopper means, not just what they typed
  • Sizing, stock and order-status questions answered instantly
  • Support team freed for the conversations that need a person

Where the build effort went (estimate)

  • AI / ML 30%
  • Backend 25%
  • Frontend 30%
  • Integrations 15%
  • Next.js
  • Shopify
  • TypeScript
  • Vector DB
  • OpenAI API
  • Stripe

Range, at a glance.

Where the build effort went on each project — an estimate, but it shows the point: she leads the AI, the backend, the product and the integrations, not just one of them.

Medical
35%30%20%15%
Fintech
25%35%20%20%
AI Calling Agent
40%25%15%20%
Real Estate
20%30%30%20%
Health & Care
15%30%35%20%
E-commerce
30%25%30%15%
AI / MLBackendFrontendIntegrations

Things she ends up explaining twice.

AI systems

Why your RAG system answers confidently and wrongly

Almost always retrieval, not the model. If the right paragraph never reaches the prompt, a good model will still answer — fluently, from nothing. Fix the chunking and the ranking before you fix the prompt, and make the system show its source so a wrong answer is visible instead of plausible.

LLM agents

An agent needs a smaller world than you think

Three narrow, typed tools beat one tool that can do anything. Every extra capability is another branch where the model can be confidently wrong, and the ones that cost money should always pause for a human. Scope is a safety feature.

Product engineering

The admin screen is the product

The demo is the client-facing page; the thing that decides whether a system survives is the screen their own team uses every day to fix what went wrong. Budget for it at the start or pay for it twice.

The three things she is brought in for.

AI agentsRAGSaaS platformsDashboardsIntegrationsMVPs

AI engineering

Agents that take an action rather than a guess, retrieval over a company’s own documents, and the evaluation that catches a regression before a user does.

  • LLM agents & tool use
  • RAG and document AI
  • Prompt & eval pipelines
  • Fine-tuning and inference

Full-stack product

The whole product: schema, auth, roles, payments, the admin screen nobody budgets for, and a front end that feels quick on a mid-range phone.

  • React & TypeScript
  • Node and Python services
  • Postgres & MongoDB
  • Auth, roles and billing

Data & integration

Getting systems that were never meant to meet to agree — and keeping them in step once they do, including on the day one of them is down.

  • REST & webhook integrations
  • ETL and data pipelines
  • Dashboards and reporting
  • CI/CD and deployment

Questions clients ask first.

What does Sohaima do at CodeXcape?

Sohaima is an AI and full-stack team lead at CodeXcape. She scopes projects, sets the architecture, leads the engineering team through delivery and is the client’s day-to-day point of contact.

Which industries has she delivered projects in?

Medical, fintech, AI calling agents, real estate, health and care, and e-commerce — from AI clinical documentation and KYC risk engines to voice agents, property platforms, home-care apps and AI shopping assistants.

Can she build an AI calling agent or voice assistant?

Yes. She has led builds of AI voice agents that answer inbound calls, run outbound follow-ups, qualify leads, book appointments into a calendar and CRM, and hand off to a human when needed.

Does she work with clients in the US, UK and Gulf?

Yes. She works remotely with teams in the US, UK and Gulf, with overlap in their working hours for calls, demos and sprint reviews.

How do I start a project with her?

Email sohaima@codexcape.solutions or reach her on Fiverr with what breaks today and what “fixed” would look like. She will reply with whether she is the right fit and a suggested first step.

Sixty seconds, in her own words.

Press play — 60 seconds

PythonTypeScriptReactNext.jsNode.jsFastAPIPostgreSQLMongoDBLangChainOpenAI APIPyTorchDockerAWSTailwindREST APIsCI/CDTwilioWhisperReact NativeStripePythonTypeScriptReactNext.jsNode.jsFastAPIPostgreSQLMongoDBLangChainOpenAI APIPyTorchDockerAWSTailwindREST APIsCI/CDTwilioWhisperReact NativeStripe

Got something hard to build?

Tell her what breaks today and what “fixed” looks like. She’ll say honestly whether she’s the right person for it.

fiverr.com/sohaachsohaima@codexcape.solutions

Pakistan — working with teams in the US, UK and Gulf