Senior Agentic AI Engineer

Location: 100% Remote – Work from Anywhere in Pakistan

Company HQ: New York, NY, USA

Contact: www.falconscaling.com | optimize@falconscaling.com | PolarityIQ.com

Compensation: 1,000,000 PKR per month starting salary

About This Opportunity

This role is for engineers who have already built and operated production-grade agentic systems under real-world conditions. We are explicitly looking for engineers who have moved beyond demos, tutorials, and isolated prototypes into production-grade systems that have survived contact with real users, operational complexity, and commercial reality. You will be asked to show us agentic builds you have already architected and you’ll be required to create a micro-agentic system to showcase your skill before being hired. If your experience with agentic systems is still primarily experimental or tutorial-based, then this role will likely not be a strong fit.

What You’ll Own

AGENTIC SYSTEMS DEVELOPMENT

This is the core of the role. Everything else supports it.

Design and build production-grade agentic systems

• Architect autonomous systems that take user goals as input, decompose them into action sequences, and execute against PolarityIQ’s data and external systems to produce structured outputs of meaningful value
• Build systems that gather, validate, enrich, and synthesize intelligence across public and proprietary sources without manual orchestration
• Design continuously operating systems that surface signals, opportunities, inconsistencies, and changes that would otherwise require manual review
• Build systems that produce outputs valuable enough to drive paid conversion, change user behavior, or replace manual workflows
• Design agents that operate with bounded autonomy rather than static prompt execution

Architect for production, not demonstration

• Build systems with explicit decision logic rather than chained prompts or scripted flows
• Design clean separation between reasoning, retrieval, tool execution, memory, and presentation layers
• Implement memory architectures that persist state across actions and sessions where the workflow requires continuity
• Build dynamic tool-selection logic that determines which tool should be used, when, and why
• Design human validation checkpoints where the system surfaces ambiguity, uncertainty, or decisions requiring judgment
• Understand the difference between a prompt chain, a workflow, a retrieval system, and a genuinely autonomous system, and know when each is appropriate
• Design for systems that survive real usage, not isolated demos

Engineer for failure, recovery, and operational reality

• Design failure handling before deployment: hallucinations, infinite loops, tool failures, degraded retrieval quality, runaway costs, rate limits, and silent failure states
• Build retry logic, fallback paths, graceful degradation, and bounded execution into every production system
• Implement observability so failures are diagnosable without manual inspection
• Build recovery patterns that allow systems to continue operating through transient failures with minimal intervention
• Design cost ceilings, time ceilings, and action ceilings so systems cannot operate unchecked
• Understand that production systems are judged by sustained reliability and usefulness, not by demo quality

Operate on PolarityIQ’s data and infrastructure

• Build on top of PolarityIQ’s retrieval, enrichment, and data infrastructure rather than rebuilding foundational layers unnecessarily
• Extend retrieval and orchestration capabilities where new workflows, data structures, or query patterns require it
• Design for real buyer workflows across family offices, institutional LPs, sovereign wealth funds, and endowments
• Build systems that support diligence, mandate matching, intelligence synthesis, outreach preparation, and relationship discovery
• Treat retrieval as a primitive inside larger decision systems rather than as the final product itself

Build foundational agentic platforms

• Architect foundational systems that support multiple client workflows, mandates, and data environments
• Design configurable platforms that adapt across client types without requiring bespoke rebuilds
• Handle variation in data sources, decision logic, output structure, and integrations through strong abstractions rather than client-specific forks
• Translate recurring operational patterns into reusable primitives that compound in value over time
• Build with the production discipline required for multi-tenant environments: isolation, observability, security, cost attribution, and extensibility
• Document systems clearly enough that internal operators, engineers, and clients can understand and extend them

Operate under real commercial constraints

• Build systems that operate under latency, reliability, infrastructure cost, and imperfect-data constraints
• Make tradeoffs between autonomy, correctness, speed, and operational cost
• Prioritize systems that create measurable leverage over systems that merely appear impressive
• Design for ambiguity, partial failure, incomplete information, and changing user behavior
• Understand that technical elegance alone is insufficient if the system does not create operational or commercial value

Build systems that compound

• Build feedback loops so systems improve from usage rather than degrade over time
• Design correction interfaces that allow human feedback to improve outputs without requiring retraining cycles
• Build evaluation harnesses against real workflows rather than synthetic benchmarks alone
• Treat each deployed system as infrastructure for the next system rather than as a one-off deliverable
• Encode learning, failure handling, and operational insight into the architecture itself

You must know the difference

• Between a chatbot and a production agentic system
• Between retrieval and autonomous decision-making
• Between a prompt chain and bounded goal-driven execution
• Between a demo and a system that survives contact with reality
• Between using AI as a feature and architecting systems around AI as a primitive
• Between agentic theater and systems that materially reduce or replace human operational work

FOUNDATIONAL CAPABILITIES

The agentic systems work above is the core of the role. The capabilities below are the foundations that work runs on. You are expected to be fluent in each of them. You are not expected to spend most of your time on them.

Machine Learning and Data Science Fluency

• Strong working knowledge of the ML and data science foundations that agentic systems rely on: entity resolution, classification, clustering, lead scoring, and predictive modeling
• Fluency with NLP pipelines for extracting structured intelligence from unstructured data, including filings, news, bios, fund documents, and web content
• Statistical rigor when interpreting model outputs and system behavior: hypothesis testing, confidence intervals, and causal inference where appropriate
• Comfort using ML models as tools inside agentic systems rather than as standalone deliverables

RAG and Retrieval Infrastructure

• Production-grade fluency with RAG architecture: ingestion, chunking, embedding selection, vector database choice, retrieval pipeline design
• Hybrid retrieval strategies combining semantic vector search with keyword matching for accuracy and relevance
• Context window management, hallucination mitigation, and retrieval evaluation as ongoing operational disciplines
• Cost-efficient scaling of retrieval systems: token economics, batching, caching, and architectural decisions that prevent runaway costs
• Treat retrieval as a primitive that agentic systems use, not as the final product

Database Architecture and Enrichment

• Fluency with relational schema design, indexing, and query performance at scale on Supabase or PostgreSQL
• Working knowledge of multi-source data enrichment using Clay, Apollo, Hunter, People Data Labs, Clearbit, and similar platforms
• Data hygiene discipline across pipelines: bounce rates, email verification, contact scoring, and stale record detection
• API integration fluency for pulling, transforming, and pushing data between systems

SaaS Technical Development

• Comfort building and maintaining features across the PolarityIQ platform when the agentic systems require new user-facing capabilities, frontend interfaces, or backend logic
• Awareness of SaaS metrics that connect technical decisions to commercial outcomes: cost per acquisition, conversion by source, churn indicators, expansion revenue

Claude Code and AI Tooling

• Operate Claude Code in terminal-based agentic workflows as a development collaborator, not as a code completer
• Apply Claude Code to multi-step pipeline orchestration, large codebase refactoring, and architectural decisions that benefit from extended context
• Daily working proficiency with Claude, ChatGPT, and Gemini for research, content, automation, and product development
• Know when to use AI tooling and when not to. The candidate who outsources their judgment to a model produces work that does not meet our standard

If any of these foundations is genuinely unfamiliar, this role will surface that quickly. You do not need to be the deepest specialist in any one of them. You need to be fluent enough that the agentic systems you build can use them as primitives without external help.

Job Type: Full-time

Pay: Rs1,000,000.00 per month

Education:

• Master's (Preferred)

Language:

• English (Preferred)

Work Location: In person