Applied AI Systems • Autonomous Operations • Technical Chief of Staff

Bridging Commercial Strategy with Autonomous Systems Execution.

I architect and deploy fault-tolerant AI systems, deterministic data pipelines, and sub-35ms decision engines to automate company workflows—without consuming core engineering sprint points. Upstream contributor to BerriAI/litellm.

View Systems Architecture Download Resume Schedule Technical Briefing
Location: Mumbai • Open to Relocation (e.g. Pune / Bangalore) & Remote • Notice: 0 Days
Zero Core Sprint Dependency

Independently builds internal tooling in Python, Node, and webhooks without pulling core devs off roadmaps

Upstream Security Contributor

Author of PR #42490 in BerriAI/litellm (In-Review: Scoped Anthropic OAuth tokens strictly to api.anthropic.com)

Token-Constrained Extraction

Dynamic Pydantic Enum schemas locked at runtime to verified client Chart of Accounts

Sub-25s Qualification Pipeline

Event-driven Express 5 microservice with SSE streaming & multi-model automated failover

Architectural Reality

Signature Engineered Systems

Working production prototypes and upstream open-source contributions engineered by Anas Wahab. Zero marketing fluff.

System 01 Token-Constrained Fiscal Extraction Engine
Repository

The Invisible Accountant

Autonomous UK Fiscal Document Extraction & HMRC Compliance Prototype

Engineered to eliminate LLM category hallucinations in automated bookkeeping. Rather than permitting models to guess account codes, this engine queries the client's live accounting Chart of Accounts and dynamically metaprograms Pydantic Enum schemas at runtime, physically restricting model output tokens to valid ledger entries.

Python 3.12 FastAPI Pydantic v2 (Dynamic Enums) PostgreSQL / asyncpg AES-256-GCM Encryption asyncio.Semaphore
  • Runtime Enum Token Constraints: Uses pydantic.create_model to dynamically synthesize enum types from live Chart of Accounts, eliminating out-of-ledger hallucinations.
  • Two-Phase Idempotency Locks: Implements atomic check-and-claim database transactions in PostgreSQL, preventing duplicate ledger entries from retrying webhooks.
  • API Concurrency Control: Uses an asyncio.Semaphore concurrency gate to throttle background external API transmissions and prevent rate-limit saturation.
agents.py (Dynamic Schema Compilation)
# Dynamic schema locked to live Chart of Accounts acct_dict = {safe_enum_name(a['Code']): a['Code'] for a in accts if a.get('Code')} tax_dict = {safe_enum_name(t['TaxType']): t['TaxType'] for t in taxes if t.get('TaxType')} DynamicAccountEnum = enum.Enum('DynamicAccountEnum', acct_dict) DynamicTaxEnum = enum.Enum('DynamicTaxEnum', tax_dict) dynamic_schema = create_model( 'DynamicExpenseCategorization', __base__=ExpenseCategorization, account_code=(DynamicAccountEnum, ...), tax_code=(DynamicTaxEnum, ...) )
System 02 Asynchronous Inbound Microservice
Repository Interactive Terminal

Layla Core Engine: Speed-to-Lead Inbound Architecture

Event-Driven Inbound Qualification & SSE Streaming Microservice

An asynchronous inbound lead interception engine engineered to compress qualification latency to sub-25 seconds. Solves API unreliability and webhook duplicate delivery through in-memory LoopGuard state deduplication and multi-model automatic failover.

Node.js / Express 5 Server-Sent Events (SSE) Multi-Model Automated Failover Stateful LoopGuard Deduplication Twilio Webhooks
  • Automated Model Failover: Real-time failover circuit breaker cascading across Gemini Flash tiers ensures continuous availability during upstream rate-limiting.
  • Stateful Lead Qualification: Evaluates commercial intent and minimum project criteria over multi-turn context before triggering scheduling workflows.
  • Stateful Deduplication Guard: Uses an in-memory lock map with flat-file persistence (loopGuard.js) to eliminate duplicate outreach triggers across retrying webhooks.
server/index.js (SSE Stream Route)
// Express 5.x: Real-Time SSE Stream Route app.post('/api/chat/web', async (req, res) => { res.setHeader('Content-Type', 'text/event-stream'); res.setHeader('Cache-Control', 'no-cache'); res.setHeader('Connection', 'keep-alive'); await streamResponse({ systemPrompt: QUALIFICATION_PROMPT, history: formattedHistory, onChunk: (chunk) => res.write(`data: ${JSON.stringify({ text: chunk })}\n\n`), onError: (err) => res.write(`data: ${JSON.stringify({ error: err.message })}\n\n`), onComplete: () => { res.write('data: [DONE]\n\n'); res.end(); } }); });
System 03 Upstream Open-Source Security Patch
BerriAI/litellm#42490

LiteLLM AI Gateway Security Patch

Multi-Tenant OAuth Credential Scope Isolation

Authored an upstream enterprise security patch in BerriAI/litellm, the leading open-source proxy for LLM routing and load-balancing. Verified across automated CodeQL, unit test suites, and regression testing pipelines.

Python LiteLLM Proxy Anthropic OAuth Enterprise Gateway CodeQL Verified
  • Vulnerability Neutralized: Scoped Anthropic OAuth credentials strictly to api.anthropic.com, eliminating multi-provider credential leakage across proxy endpoints.
  • Multi-Tenant Hardening: Ensures enterprise proxy instances managing shared gateways cannot inadvertently transmit OAuth bearer tokens to auxiliary endpoints.
  • Status: Authored by Anas Wahab; currently In-Review in upstream BerriAI/litellm (PR #42490, closes issue #42172).
litellm/llms/anthropic/common_utils.py
# Upstream Security Patch (PR #42490 by anaswahab231997) def optionally_handle_anthropic_oauth(headers: dict, api_key: str | None, api_base: str | None = None): auth_header = next((v for k, v in headers.items() if k.lower() == "authorization"), "") def _is_anthropic_host(url: str | None) -> bool: if not url: return True try: return urlparse(url).hostname == "api.anthropic.com" except ValueError: return False if not _is_anthropic_host(api_base): # Strip Anthropic OAuth when targeting custom / non-Anthropic endpoints if auth_header.startswith(f"Bearer {ANTHROPIC_OAUTH_TOKEN_PREFIX}"): headers = {k: v for k, v in headers.items() if k.lower() != "authorization"} if api_key and api_key.startswith(ANTHROPIC_OAUTH_TOKEN_PREFIX): api_key = None return headers, api_key
System 04 Hybrid Decision Architecture (System 1 + System 2 Orchestration)
NandhaKishorM/laya ~33ms CPU Forward Pass

Hybrid Intent Triage & Generative Routing Blueprint

Architectural Integration: Laya (by NandhaKishorM) + Dynamic Fallback Gateways

Standard industry workflows route simple operational decisions through multi-billion-parameter LLMs, incurring 1.5–3.0s latency, unpredictable schema formatting, and hallucination risks. My systems architecture integrates Laya—an open-source non-autoregressive encoder decision model created by Nandakishor Mukkunnoth—as the deterministic System 1 layer before delegating to generative models.

Laya Decision Engine ModernBERT / mmBERT ~33ms CPU Forward Pass Zero Generative Hallucination Typed Choice Constraints Gemini / Claude Fallback
  • System 1 (Deterministic Triage): Laya classifies input intent into typed choices (choice, score, noul) within ~33ms on CPU/edge without generating tokens—mathematically eliminating text hallucination.
  • System 2 (Deep Synthesis): Only when high-dimensional extraction or complex reasoning is required does the engine dispatch to Gemini or Claude via resilient rate-limited gateways.
  • Production Economics: Filters out routine classification volume before touching LLM endpoints, slashing API overhead and delivering instantaneous workflow reactions.
integration_pattern_laya.py (Architectural Blueprint)
# System 1: Deterministic Intent Gate (~33ms) decision = laya.decide( context=event.payload, choices=["deflect_faq", "escalate_human", "deep_reason"] ) if decision.choice != "deep_reason": # Zero token generation, zero hallucination return execute_deterministic_sop(decision.choice) # System 2: Generative Reasoning (Gemini/Claude) return await generative_gateway.extract(event.payload)
Deployable AI Solutions

Applied AI Workflows & Operations Architecture

Universal operational workflows engineered to automate high-friction touchpoints, deflect repetitive support volume, and eliminate data silos across modern tech and e-commerce ecosystems.

Support Operations

Automated Support Ticket Triage & Deflection Engine

Customer support queues are routinely overwhelmed by recurring interface issues, configuration errors, and routine FAQs that drain senior agent hours.

Proposed System: Headless DOM and API inspection pipeline coupled with sub-50ms intent classification. Automatically diagnoses customer storefront/interface state, isolates root anomalies, and compiles verified remediation snippets before the ticket reaches a human agent.

Headless DOM Inspector Laya Decision Routing L1 Deflection Engine
Revenue Retention

Proactive Retention & Churn Anomaly Telemetry

Protecting recurring SaaS and subscription revenue by detecting subtle usage drop-offs, broken API webhooks, or configuration errors weeks before an account churns.

Proposed System: Real-time telemetry ingestion pipeline evaluating event frequencies, billing status changes, and webhook anomalies. Triggers proactive Slack alerts and automated account remediation playbooks 14 days prior to contract renewal cycles.

Webhook Telemetry Anomaly Detection Automated Escalation
Data Orchestration

Cross-Platform Webhook & Data Sync Orchestrator

Disjointed operational tools (CRMs, payment gateways, ERPs, inventory systems) frequently desynchronize, causing silent failures, duplicate records, and manual data-entry taxes.

Proposed System: Distributed, event-driven integration layer built with PostgreSQL transactional check-and-claim locks and in-memory LoopGuard state for atomic idempotency, schema enforcement via dynamic Pydantic models, and automated dead-letter queues (DLQ) for zero data loss.

Two-Phase Idempotency Dead-Letter Queues Atomic Data Sync
Execution Framework

Production AI Workflow Deployment Blueprint

How I hit the ground running with teams and founders—deploying zero-engineering-tax automation and turning operational bottlenecks into autonomous systems.

Phase 1: Discovery & Bottleneck Audit

Shadow operations, customer success, and data intake. Map manual repetition, quantify ticket friction points, and identify high-ROI automation targets with zero engineering disruption.

Phase 2: Hybrid System 1/2 Automation

Deploy rapid sub-50ms intent routing for triage, wire schema-constrained document extraction flows, and establish robust event-driven webhook synchronization with error budgets.

Phase 3: Autonomous Scale & SOP Handover

Standardize operating SOPs, integrate automated telemetry dashboards, and transition workflows into self-healing, auditable infrastructure engineered to eliminate recurring manual triage drag.

Schedule Technical Deployment Briefing Download Blueprint (PDF) Inspect System Architectures
Execution Stack

The Operator's Arsenal

Tools and technologies I use to build internal leverage, automate operational bottlenecks, and analyze business telemetry.

AI Systems & Gateways

Production LLM routing, token security, and deterministic schema extraction.

Laya Decision Engine (ModernBERT) LiteLLM Proxy Google Gemini (Flash & Pro) Anthropic Claude Pydantic v2 (Dynamic Enums) Structured Outputs

Backend & Concurrency

Asynchronous microservices, rate-limiting, and distributed locking.

Python 3.12 (FastAPI) Node.js (Express 5.x) asyncpg / PostgreSQL Stateful Deduplication & LoopGuard Server-Sent Events

Shopify & E-Commerce

Storefront operations, app ecosystems, and merchant telemetry.

Shopify Admin APIs Liquid & Variant Architecture App Store Dynamics Funnel Economics (CAC/LTV) Theme Triage

Automation & Internal Ops

Event-driven workflows, data syncing, and cross-functional alerting.

n8n (Queue Mode) Webhook Ingestion Supabase Edge Functions Slack API / HubSpot Puppeteer Headless
Verified Track Record

Operational & Commercial Leadership

7 years bridging commercial strategy, e-commerce supply chains, and autonomous systems execution. Grounded in P&L ownership, operational leverage, and measurable business impact.

2024 – PRESENT

Applied AI Systems Operator & Technical Architect

Nexlify Systems Laboratory • Independent Systems Engineering
Applied AI Systems

Architecting and deploying production AI systems, deterministic schema engines, and event-driven microservices to automate company operations with zero developer tax on core engineering teams.

  • Upstream Contributor: Authored security patch PR #42490 in BerriAI/litellm, scoping Anthropic OAuth bearer tokens strictly to authorized hosts.
  • Deterministic Tax Extraction: Architected The Invisible Accountant, using dynamic Pydantic Enum metaprogramming locked to live Chart of Accounts and AES-256-GCM encryption.
  • Asynchronous Speed-to-Lead: Engineered Layla Core, an Express 5 SSE microservice with in-memory LoopGuard state deduplication achieving sub-25s response times.
SEP 2023 – PRESENT

Founder & Director of Operations

Cashmerean • Luxury E-Commerce & Cross-Border Supply Chain
E-Commerce & Ops

Direct ownership of end-to-end e-commerce supply chain, Shopify merchant operations, international artisan coordination, and unit economics optimization.

  • Shopify Architecture & Variant Mastery: Managed Shopify catalog structure, product variant architectures, inventory tracking, and custom storefront workflows.
  • Supply Chain & Margin Protection: Orchestrated cross-border logistics between Kashmir artisan cooperatives and global luxury buyers with zero inventory leakage.
  • Operational Automation: Engineered automated vendor dispatching, tracking updates, and customer communications using webhook-driven pipelines.
MAY 2019 – MAY 2023

Senior Digital Marketing Head & Growth Lead

Content Garage • Growth & Performance Marketing
Growth & Analytics

Led digital performance marketing, acquisition funnel architectures, and analytics infrastructure across high-growth brand campaigns.

  • 22% CAC Reduction: Re-engineered conversion funnel attribution models and campaign mechanics, delivering a sustained 22% reduction in customer acquisition cost.
  • Cross-Functional Leadership: Managed creative, media buying, and analytics teams; aligned multi-stakeholder roadmaps with executive commercial targets.
  • Performance Telemetry: Built reporting dashboards and tracking telemetry to provide real-time CAC/LTV visibility for executive leadership.
2016 – 2019

Bachelor of Mass Media (BMM)

Thakur College of Science & Commerce • University of Mumbai
Executive Foundation

Specialized in strategic communication, media economics, and organizational persuasion. Provides the high-leverage narrative and executive writing foundation essential for a CEO-facing Chief of Staff.

  • Executive Alignment: Trained in clear, concise stakeholder communication and cross-functional leadership messaging.
  • Analytical Grounding: Foundation in media economics, audience segmentation, and quantitative research methods.
Profile Dossier • Applied AI Systems Operator & Technical Chief of Staff

Anas Wahab

I am a technical operator bridging commercial strategy, e-commerce operations, and autonomous AI systems. I hold a Bachelor's in Mass Media (Thakur College, University of Mumbai, 2016–2019), giving me strong stakeholder alignment, strategic communication, and executive writing capabilities.

My operational foundation is built on 4 years leading performance marketing at Content Garage (cutting CAC by 22%), followed by building and operating Cashmerean, where I managed international luxury supply chains, cross-border fulfillment, and Shopify variant catalogs.

Unlike traditional chiefs of staff who only build slide decks, I build internal software and automated pipelines myself. Through independent systems engineering (Nexlify Systems Laboratory), I build production-grade prototypes, deterministic schema engines, and event-driven automation architectures to solve operational bottlenecks without burdening core dev teams.

I am prepared to dedicate 100% of my operational focus as an in-house technical systems builder and executive force multiplier. Based in Mumbai, with full flexibility for immediate relocation (e.g. Pune, Bangalore) or remote deployment with 0 days notice.

View Resume (PDF) GitHub LinkedIn anaswahabofficial@gmail.com