Initiative: Sentinel Ledger — Real-Time FinTech Risk Intelligence & Anomaly Surveillance Platform (Backend Engineering)
Budget / Salary£250–750
TypeFreelance project
LocationRemote
Posted2 hours ago
*Initiative: Sentinel Ledger — Real-Time FinTech Risk Intelligence & Anomaly Surveillance Platform (Backend Engineering)*
We are seeking a senior systems architect or technical team to design, build, containerize, and operate the production backend for a real-time anomaly detection and threat-scoring platform for financial transactions.
This is not a proof-of-concept or an experimental sandbox. The final deliverable must be a fully hardened, real-time, fault-tolerant system built to run in live enterprise environments.
*System Overview*
The backend framework must:
- Ingest and process high-velocity transaction data continuously and without lag
- Detect irregular behavior using trained machine learning models, not static or hardcoded heuristics
- Publish live threat-assessment metrics through fast, low-latency endpoints
- Meet demanding throughput, response-time, and scaling requirements
Placeholder code, hardcoded logic, synthetic mock data, or simulated streaming endpoints will be treated as an automatic project default.
*Compensation & Milestone Schedule — GBP 2,300 Fixed, Paid in Two Stages*
_Stage 1 — GBP 2,185_
Released when the contract begins. Covers cloud resources, environment setup, and the initial core pipeline build.
Note: this payment is not delivery, review, or approval of the work. It does not waive any technical spec, throughput standard, or runtime target, and it does not limit legal recourse or clawback rights if the contract is breached.
_Stage 2 — GBP 115 (Final Settlement)_
Released only once:
- Production go-live is complete, end to end
- All microservices are active and reachable
- Latency and throughput metrics have been verified
- Full handover has taken place: source repos, model weights/artifacts, cloud/infra credentials, technical/architecture docs
If exit criteria aren't met, final payment is forfeited.
*What Needs to Get Built*
Leave out any piece below and the whole delivery counts as failed.
_Data Intake_
- Distributed streaming, running on Kafka and Spark Streaming
- Handles serious payload volume without falling behind, in real time
- Schema checks, cleanup, and validation baked into the pipeline itself
_Feature Engineering_
- One central place where computed features live
- Real-time streaming and offline batch computation running side by side
_Detection & Scoring_
- Anomaly detection that mixes statistical boundaries with recurrent LSTM models
- A weighted service that blends multiple models into one threat score
- Model inference that returns results in well under a second
- Fixed rule filters, hardcoded thresholds, or canned static responses won't cut it here.
_Making Predictions Explainable_
- SHAP-based explainability (or something equivalent)
- A plain-language reason attached to every score the model produces
_API & Application Layer_
- Built on FastAPI, organized into clean modules
- OAuth2/JWT handles enterprise authentication
- Structured to run as standalone microservices
- Complete OpenAPI/Swagger docs for anyone consuming the API downstream
_Infrastructure & DevOps_
- Kubernetes running across multiple nodes
- CI/CD automated end to end for builds and releases
- Grafana and Prometheus covering telemetry, logging, and observability
- Canary or blue-green deploys for safe rollouts
- A script that only runs locally, a docker-compose setup with no real scaling path, or anything that can't handle production traffic gets disqualified.
_Testing & QA_
- Load and concurrency testing pushed to peak volume
- A formal readiness check signed off before anything goes live
*Mandatory Performance Benchmarks*
The architecture must meet or exceed every one of these:
- Ingestion Rate: ≥ 50,000 events/sec
- Pipeline Latency: ≤ 2.0 seconds
- Feature Lookup: ≤ 50 milliseconds
- Model Precision: ≥ 90% true-positive anomaly rate
- Inference Compute Time: ≤ 80 milliseconds
- API Round-Trip Response: ≤ 400 milliseconds
Missing a single benchmark constitutes non-delivery.
*Uptime & Reliability*
99.5% uptime, tracked monthly. Usual carve-outs apply — scheduled maintenance, third-party vendor outages, force majeure, or unauthorized changes.
*Support Coverage*
Hours: 09:00–18:00 GMT/BST
- Sev 1 (everything's down): triage within 4 hrs, patched within 24 hrs
- Sev 2 (major functionality broken): triage within 8 hrs, patched within 48 hrs
- Sev 3 (small issue): triage within 1 business day, resolved within 5
*When Is This Actually Done?*
Sign-off happens only when, all at once:
- It's live and running in a real cloud environment
- Every listed component is up and checked
- The models are giving real predictions, not sample output
- Latency and throughput hold up under actual load
- CI/CD and monitoring are working
- Architecture docs and runbooks delivered
- Credentials, repos, and model files fully handed over
Half-finished doesn't count as finished here.
*Security, Governance & Regulatory Compliance*
Must follow safeguards aligned with ISO/IEC 27001, SOC 2, and applicable PCI-DSS standards. All pipelines must comply with UK/EU GDPR. Anonymized, non-attributable data may be used only where legally permitted.
*Intellectual Property Rights*
Once the final milestone is released, all custom code, model weights, config scripts, and documentation transfer entirely to the Employer. Developer keeps prior rights to their own internal boilerplate, generic scaffolding, and reusable provisioning utilities. Employer retains exclusive ownership of all proprietary data and live streams.
*Non-Disclosure & Confidentiality*
Both parties keep confidential: platform architecture, data workflows/models, live payloads, secrets, and system configs. This obligation stays in force indefinitely after the contract ends.
*Default, Breach & Recourse*
Failing to deliver a fully tested, production-grade platform is a material breach — missed deadlines, broken/incomplete services, unmet performance thresholds, or inability to run live in production. Employer can cancel the engagement and seek full restitution of paid funds under governing law.
*Right to Suspend Work*
Developer may pause work if invoices go unpaid beyond 30 days, Employer breaches core terms, misuse/tampering is uncovered, or an imminent security vulnerability threatens the build.
*Liability Bounds*
Standard statutory liability limits apply. Nothing here limits liability for death/personal injury from negligence, deliberate fraud/misrepresentation, or non-waivable statutory duties.
*Unforeseen Events (Force Majeure)*
Neither party is responsible for delays from events outside reasonable control: natural disasters, armed conflict, grid/telecom blackouts, state-level emergencies or epidemics.
*Communication & Governance Channels*
All official communication must go through Upwork platform channels. Off-platform commitments carry no contractual weight.
*Legal Framework*
Governed exclusively by the laws of England and Wales; disputes subject to the sole jurisdiction of its courts.
*Zero Partial-Acceptance Policy*
Piecemeal or fractional work won't be accepted. No partial payment for semi-functional builds. "Near complete," "pre-alpha," or "beta-grade" submissions rejected. No prorated payouts for individual working modules. Test-suite access or interim feedback ≠ formal acceptance. Acceptance is binary and requires 100% production completion.
*Candidate Prerequisites*
- Proven track record shipping high-availability distributed backends to production
- Hands-on mastery of high-throughput streaming architectures
- Experience containerizing and serving scalable ML systems
- Deep operational fluency with Kafka, Spark, Kubernetes, and FastAPI
*Submission Checklist*
- Case studies of live production builds (not demos/sandboxes)
- High-level technical design proposal
- Team layout/resource allocation (if bidding as an agency)
- Milestone-based delivery timeline
*Final Directive*
This engagement requires enterprise-tier systems engineering. Anything short of a fully deployed, high-throughput, production-ready platform will be marked as non-performance and unfulfilled delivery.
We are seeking a senior systems architect or technical team to design, build, containerize, and operate the production backend for a real-time anomaly detection and threat-scoring platform for financial transactions.
This is not a proof-of-concept or an experimental sandbox. The final deliverable must be a fully hardened, real-time, fault-tolerant system built to run in live enterprise environments.
*System Overview*
The backend framework must:
- Ingest and process high-velocity transaction data continuously and without lag
- Detect irregular behavior using trained machine learning models, not static or hardcoded heuristics
- Publish live threat-assessment metrics through fast, low-latency endpoints
- Meet demanding throughput, response-time, and scaling requirements
Placeholder code, hardcoded logic, synthetic mock data, or simulated streaming endpoints will be treated as an automatic project default.
*Compensation & Milestone Schedule — GBP 2,300 Fixed, Paid in Two Stages*
_Stage 1 — GBP 2,185_
Released when the contract begins. Covers cloud resources, environment setup, and the initial core pipeline build.
Note: this payment is not delivery, review, or approval of the work. It does not waive any technical spec, throughput standard, or runtime target, and it does not limit legal recourse or clawback rights if the contract is breached.
_Stage 2 — GBP 115 (Final Settlement)_
Released only once:
- Production go-live is complete, end to end
- All microservices are active and reachable
- Latency and throughput metrics have been verified
- Full handover has taken place: source repos, model weights/artifacts, cloud/infra credentials, technical/architecture docs
If exit criteria aren't met, final payment is forfeited.
*What Needs to Get Built*
Leave out any piece below and the whole delivery counts as failed.
_Data Intake_
- Distributed streaming, running on Kafka and Spark Streaming
- Handles serious payload volume without falling behind, in real time
- Schema checks, cleanup, and validation baked into the pipeline itself
_Feature Engineering_
- One central place where computed features live
- Real-time streaming and offline batch computation running side by side
_Detection & Scoring_
- Anomaly detection that mixes statistical boundaries with recurrent LSTM models
- A weighted service that blends multiple models into one threat score
- Model inference that returns results in well under a second
- Fixed rule filters, hardcoded thresholds, or canned static responses won't cut it here.
_Making Predictions Explainable_
- SHAP-based explainability (or something equivalent)
- A plain-language reason attached to every score the model produces
_API & Application Layer_
- Built on FastAPI, organized into clean modules
- OAuth2/JWT handles enterprise authentication
- Structured to run as standalone microservices
- Complete OpenAPI/Swagger docs for anyone consuming the API downstream
_Infrastructure & DevOps_
- Kubernetes running across multiple nodes
- CI/CD automated end to end for builds and releases
- Grafana and Prometheus covering telemetry, logging, and observability
- Canary or blue-green deploys for safe rollouts
- A script that only runs locally, a docker-compose setup with no real scaling path, or anything that can't handle production traffic gets disqualified.
_Testing & QA_
- Load and concurrency testing pushed to peak volume
- A formal readiness check signed off before anything goes live
*Mandatory Performance Benchmarks*
The architecture must meet or exceed every one of these:
- Ingestion Rate: ≥ 50,000 events/sec
- Pipeline Latency: ≤ 2.0 seconds
- Feature Lookup: ≤ 50 milliseconds
- Model Precision: ≥ 90% true-positive anomaly rate
- Inference Compute Time: ≤ 80 milliseconds
- API Round-Trip Response: ≤ 400 milliseconds
Missing a single benchmark constitutes non-delivery.
*Uptime & Reliability*
99.5% uptime, tracked monthly. Usual carve-outs apply — scheduled maintenance, third-party vendor outages, force majeure, or unauthorized changes.
*Support Coverage*
Hours: 09:00–18:00 GMT/BST
- Sev 1 (everything's down): triage within 4 hrs, patched within 24 hrs
- Sev 2 (major functionality broken): triage within 8 hrs, patched within 48 hrs
- Sev 3 (small issue): triage within 1 business day, resolved within 5
*When Is This Actually Done?*
Sign-off happens only when, all at once:
- It's live and running in a real cloud environment
- Every listed component is up and checked
- The models are giving real predictions, not sample output
- Latency and throughput hold up under actual load
- CI/CD and monitoring are working
- Architecture docs and runbooks delivered
- Credentials, repos, and model files fully handed over
Half-finished doesn't count as finished here.
*Security, Governance & Regulatory Compliance*
Must follow safeguards aligned with ISO/IEC 27001, SOC 2, and applicable PCI-DSS standards. All pipelines must comply with UK/EU GDPR. Anonymized, non-attributable data may be used only where legally permitted.
*Intellectual Property Rights*
Once the final milestone is released, all custom code, model weights, config scripts, and documentation transfer entirely to the Employer. Developer keeps prior rights to their own internal boilerplate, generic scaffolding, and reusable provisioning utilities. Employer retains exclusive ownership of all proprietary data and live streams.
*Non-Disclosure & Confidentiality*
Both parties keep confidential: platform architecture, data workflows/models, live payloads, secrets, and system configs. This obligation stays in force indefinitely after the contract ends.
*Default, Breach & Recourse*
Failing to deliver a fully tested, production-grade platform is a material breach — missed deadlines, broken/incomplete services, unmet performance thresholds, or inability to run live in production. Employer can cancel the engagement and seek full restitution of paid funds under governing law.
*Right to Suspend Work*
Developer may pause work if invoices go unpaid beyond 30 days, Employer breaches core terms, misuse/tampering is uncovered, or an imminent security vulnerability threatens the build.
*Liability Bounds*
Standard statutory liability limits apply. Nothing here limits liability for death/personal injury from negligence, deliberate fraud/misrepresentation, or non-waivable statutory duties.
*Unforeseen Events (Force Majeure)*
Neither party is responsible for delays from events outside reasonable control: natural disasters, armed conflict, grid/telecom blackouts, state-level emergencies or epidemics.
*Communication & Governance Channels*
All official communication must go through Upwork platform channels. Off-platform commitments carry no contractual weight.
*Legal Framework*
Governed exclusively by the laws of England and Wales; disputes subject to the sole jurisdiction of its courts.
*Zero Partial-Acceptance Policy*
Piecemeal or fractional work won't be accepted. No partial payment for semi-functional builds. "Near complete," "pre-alpha," or "beta-grade" submissions rejected. No prorated payouts for individual working modules. Test-suite access or interim feedback ≠ formal acceptance. Acceptance is binary and requires 100% production completion.
*Candidate Prerequisites*
- Proven track record shipping high-availability distributed backends to production
- Hands-on mastery of high-throughput streaming architectures
- Experience containerizing and serving scalable ML systems
- Deep operational fluency with Kafka, Spark, Kubernetes, and FastAPI
*Submission Checklist*
- Case studies of live production builds (not demos/sandboxes)
- High-level technical design proposal
- Team layout/resource allocation (if bidding as an agency)
- Milestone-based delivery timeline
*Final Directive*
This engagement requires enterprise-tier systems engineering. Anything short of a fully deployed, high-throughput, production-ready platform will be marked as non-performance and unfulfilled delivery.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.