Staff Software Engineer, Applied AI Systems

Hansal Maniar

Staff Software Engineer

Ontario, Canada Remote +1-289-***-**** ******.******@*****.*** https://www.linkedin.com/in/humaniar/

SUMMARY

Staff Software Engineer who builds and operationalizes applied-AI systems that combine retrieval-

augmented generation, agent tool-use and document-aware pipelines to deliver production features. Leads

end-to-end delivery from prototype to rollout, embedding observability, regression tests and rollback

procedures into services and inference platforms. Mentors engineers and codifies reusable patterns, libraries

and runbooks that accelerate safe, auditable deployment of LLM-based features.

E XPERIENCE

Wealthsimple Staff Software Engineer (Aug 2025 – Present)

Remote

• Owned end-to-end design and shipping of generative-AI features inside the Book of Record platform,

integrating RAG retrieval and agentic controls with the Financial Activity Model and Financial Calculation

Platform using OpenAI models and vector search with pgvector in PostgreSQL.

• Architected a document-extraction pipeline for leases and contracts that combined layout-aware OCR,

bespoke parsers and human-in-the-loop review, and integrated extracted entities into the Machine Learning

Platform for downstream scoring.

• Implemented an inference gateway that routed requests to OpenAI, applied prompt/version control, and

enforced guardrails and permissioning at the API layer using role-based access, reducing unsafe model calls

by 22 percent.

• Built an observability and regression-testing stack for model and prompt changes using structured logging,

OpenTelemetry traces and datadog-compatible metrics, and automated rollout gating during canary releases.

• Led the streaming recomputation work for the Financial Calculation Platform, implementing Kafka consumers

and idempotent processors that kept materialized views consistent with the single write funnel, supporting 1.8

million daily financial events.

• Implemented production-ready rollback and versioning procedures for agent behavior, storing agent policy

and prompt versions in a model registry and enabling service-level rollback within 9 minutes during incidents.

• Developed scalable vector search backed by pgvector and Redis for fast similarity lookups in customer

documents and policy text, reducing median retrieval latency to 85 milliseconds for the primary read path.

• Built REST APIs and gRPC services that exposed agent capabilities to the app and internal tools, instrumented

with request tracing and SLOs tied to the Financial Activity Model write funnel.

• Prototyped and evaluated multiple foundation model candidates against operational documents and

telemetry using systematic evaluation harnesses, selecting OpenAI as the primary provider based on cost,

latency and safety criteria.

• Codified applied-AI patterns and libraries for prompt engineering, structured JSON-mode outputs, and agent

tool invocation, and mentored product engineers to adopt the libraries across the Book of Record estate.

TEEMA Senior Software Engineer (Nov 2020 – Jul 2025)

Winnipeg, Manitoba, Canada Remote

• Built a multi-tenant replatforming for a B2B SaaS startup client, delivering tenant isolation, per-tenant

configuration, rate-limited public APIs and an admin analytics surface deployed to Azure and secured with

enterprise SSO.

• Implemented per-tenant observability and SLO-backed monitoring on the replatformed API gateway,

instrumenting traces and metrics that exposed cross-tenant noisy-neighbor incidents and reduced noisy-

neighbor occurrences by 18 percent.

• Rewrote the ingestion and metering pipeline to feed billing and entitlements, using Kafka for event streaming

and Azure Functions for fan-out processing, enabling accurate usage billing across 3,200 tenants.

• Delivered a versioned GraphQL/REST API surface with backward-compatible schema evolution and blue-green

deploys managed by Terraform and an Azure CI pipeline, cutting rollback time from 40 minutes to 13 minutes.

• Delivered a routing and optimization system for a logistics startup client, building real-time tracking ingestion

and a constraint-based routing engine with forecasting models for demand and ETA.

• Built an embeddings-based similarity layer using FAISS and Pinecone for shipment and route retrieval,

connected to LangChain flows and LlamaIndex utilities to support multimodal lookup of manifests and proof-

of-delivery images, serving 42,000 queries per day.

• Implemented document-AI pilots that combined OCR, layout extraction and human-review gates for bills of

lading, and wired review outcomes back into model retraining pipelines to improve structured extraction

accuracy.

• Integrated Claude as a model provider for agentic coordination tasks, implementing tool-calling agents with

strict action risk controls, action logging and permission checks before any downstream system call.

• Created reusable runbooks, CI templates and component libraries that packaged RAG, vector store integration

and guardrail patterns, enabling client engineering teams to adopt the pilots into their own Azure

deployments.

• Built regression testing harnesses and drift monitoring that computed end-to-end behavior deltas after

prompt and model updates, and configured alerts when behavioral divergence exceeded defined thresholds.

Suncore Construction & Materials Senior Software Engineer (Aug 2018 – Oct 2020)

Remote

• Built backend services and integrations that connected the cloud CRM used by sales and estimating with

dispatch, telematics and job-cost systems, replacing manual re-keying between platforms that had grown up

separately.

• Delivered quoting and project workflows for ready-mix and construction operations, surfacing customer

history and live operational data together so sales teams could price and commit work without leaving the

application.

• Developed data pipelines consolidating project, job-cost, materials and fleet telemetry into reporting models

that gave operations and finance one reconcilable view across business units.

• Built mobile-accessible field workflows and the APIs behind them so crews and dispatchers could record and

retrieve operational data on site rather than routing everything through the office.

Google Senior Software Engineer (Sep 2013 – Jun 2018)

Toronto, Ontario, Canada

• Built backend components in Google Ads serving, developing Java and C++ services that applied advertiser

budget, pacing and eligibility constraints during latency-sensitive ad selection across global traffic.

• Developed FlumeJava and MapReduce pipelines that aggregated impression, click and conversion events into

advertiser-spend, budget-pacing and reporting datasets used by campaign-management workflows.

• Built reconciliation and correctness checks across the event-to-spend path so logged impressions, recorded

clicks and spend calculations stayed consistent under retries, delayed events and pipeline reprocessing.

• Integrated machine-learned relevance and conversion signals into production ranking and pacing decisions,

moving features from offline evaluation into controlled online experiments without regressing serving latency.

• Contributed to production readiness and shared on-call rotations for Ads services, tracing spend and serving

regressions across dependent systems and working with infrastructure and data teams on permanent fixes.

CIBC Application Developer (Jun 2007 – Aug 2013)

Toronto, Ontario, Canada

• Built client-facing banking applications and the digital-channel services behind them as CIBC extended from

online banking into mobile, covering account access, transfers, bill payment and person-to-person money

movement.

• Developed integration layers between customer applications and core banking systems, isolating channel

development from mainframe release cycles so new digital-channel features could evolve without waiting on

core-system release cycles.

• Implemented authentication, authorization and transaction security controls for a channel that was new to the

bank, working within the audit and regulatory requirements of Canadian retail banking.

• Delivered payment and transaction workflows with the availability and correctness guarantees money

movement demands, including reconciliation and failure handling on client-initiated transfers.

SKILLS

Generative & Agentic AI: LLMs, RAG, LangChain, LlamaIndex, embeddings, vector search, prompt

engineering, prompt/versioning, guardrails, LLM evaluation

AI & Machine Learning: Python, PyTorch, TensorFlow, scikit-learn, XGBoost, pandas, NumPy, feature

engineering, model evaluation, anomaly detection

MLOps & Model Systems: MLflow, Kubeflow, Amazon SageMaker, Azure ML, Triton Inference Server, ONNX,

model serving, experiment tracking, model registries, drift monitoring

Languages & Backend: Python, TypeScript, Java, Go, SQL, JavaScript, FastAPI, REST, gRPC, GraphQL

Databases, Search & Retrieval: PostgreSQL, pgvector, Pinecone, FAISS, Elasticsearch/OpenSearch, Redis,

MongoDB, indexing and query optimization

Cloud & Platform: AWS, Azure, GCP, EC2, S3, Lambda, EKS, Azure Functions, Terraform, API Gateway

Observability & Reliability: OpenTelemetry, Prometheus, Grafana, Datadog, Jaeger, structured logging,

distributed tracing, SLOs/SLAs, regression testing for model changes

EDUCATION

Bapuji Institute of Engineering & Technology, Davangere, Master, Computer Science

Contact this candidate