senior-backend-engineer-ai-backend-lead@anaqua:~/career
← Back to CV
Acquired Enterprise AI Lead

Senior Backend Engineer & AI Backend Lead

Anaqua 🇺🇸 Boston, Massachusetts
📅 September 2023 → December 2025 (2 years 4 months)

$ echo $IMPACT_METRICS

50% Faster Search
99.9% System Uptime
0→1 AI Backend Built

$ cat tech-stack.json

🤖 AI & Machine Learning

OpenAI GPT-4Anthropic ClaudeGoogle GeminiHuggingFaceLangChainRAG PipelinesCAG (Context-Aware Generation)AI AgentsModel Context Protocol (MCP)

⚡ Core Technologies

JavaSpring BootPythonFastAPITypeScript

🔧 Supporting Stack

PostgreSQLPGVectorJSONBReactRedux

☁️ Infrastructure & DevOps

GCPAWSDockerKubernetesGitLab CI/CD

$ cat README.md

At RightHub (acquired by Anaqua), I was brought in to build the entire AI backend from the ground up. This wasn’t about adding AI features to an existing product — it was about fundamentally transforming how the platform handled knowledge, search, and decision-making.

The challenge was significant: take a successful IP management platform and infuse it with generative AI capabilities that enterprise customers could trust with their sensitive intellectual property data.

Over 2+ years, I architected and shipped systems that are now processing thousands of daily queries, enabling legal and IP professionals to work 50% faster while maintaining the strict compliance requirements of enterprise software.

$ git log --oneline responsibilities/

→
Architected the complete AI backend infrastructure from scratch, establishing patterns and practices that the entire engineering team adopted
→
Designed and implemented RAG (Retrieval-Augmented Generation) systems that enable semantic search across millions of IP documents, patents, and legal filings
→
Built autonomous AI agents using LangChain and custom frameworks that can analyze documents, extract entities, and make recommendations with human oversight
→
Developed Context-Aware Generation (CAG) pipelines that leverage prompt chaining, session memory, and user/task metadata for highly personalized AI experiences
→
Integrated multiple LLM providers (OpenAI, Anthropic, Gemini, HuggingFace) with intelligent routing based on task complexity and cost optimization
→
Implemented vector search infrastructure using PGVector with custom embedding pipelines optimized for legal and IP terminology
→
Established Model Context Protocol (MCP) integrations enabling AI agents to safely interact with external legal databases and patent offices
→
Led cross-functional collaboration with product, design, and frontend teams to embed AI-driven features across the enterprise platform

$ grep -r "achievement" ./

✓
Spearheaded the AI backend that transformed the company’s knowledge and data systems, processing 10,000+ daily AI-powered queries
✓
Achieved 50% faster search through vector-based RAG systems, reducing time for IP professionals to find relevant documents from minutes to seconds
✓
Built and deployed autonomous AI agents that integrate real-time data with planning algorithms, automating complex IP analysis workflows
✓
Maintained 99.9% uptime across all AI services with elastic scalability handling 10x traffic spikes during product launches
✓
Reduced LLM costs by 40% through intelligent prompt caching, model routing, and response compression strategies
✓
Pioneered CAG (Context-Aware Generation) methodology that became a core differentiator for the product’s AI capabilities

$ cat CHALLENGES.md

Enterprise-Grade RAG for Legal Documents

🔴 Challenge:

Legal and patent documents contain highly specialized terminology, complex document structures, and citation networks that generic RAG approaches couldn't handle effectively.

🟢 Solution:

Developed a custom chunking strategy that respects document structure (claims, citations, legal provisions). Built a specialized embedding model fine-tuned on IP/legal corpus. Implemented citation-aware retrieval that follows reference chains to provide complete context.

PGVectorLangChainCustom EmbeddingsPostgreSQL JSONB

Multi-LLM Orchestration with Cost Control

🔴 Challenge:

Different AI tasks required different model capabilities, but naive approaches led to either poor quality (using cheap models for complex tasks) or exploding costs (using GPT-4 for everything).

🟢 Solution:

Built an intelligent routing layer that classifies task complexity in real-time and routes to appropriate models. Implemented prompt caching with semantic similarity matching, reducing redundant LLM calls by 60%. Added circuit breakers and fallback chains for reliability.

PythonFastAPIRedisLangChainCustom Router

Reliable AI Agents for Enterprise Workflows

🔴 Challenge:

AI agents needed to perform multi-step tasks (document analysis, entity extraction, comparison) but would occasionally hallucinate or make incorrect tool calls that could corrupt data.

🟢 Solution:

Implemented structured output validation using Pydantic models for every agent action. Built a human-in-the-loop approval system for high-stakes operations. Added comprehensive observability with LangSmith and custom metrics to detect anomalies early.

LangGraphPydanticLangSmithPostgreSQLReact

Context-Aware Generation at Scale

🔴 Challenge:

Generic AI responses weren't personalized enough for enterprise users who needed context-aware assistance based on their role, current project, and historical interactions.

🟢 Solution:

Designed CAG (Context-Aware Generation) architecture that maintains session memory, injects user/task metadata into prompts, and implements prompt chaining for multi-turn conversations. This became a core product differentiator.

RedisPostgreSQLLangChainCustom Prompting Framework

$ cat details.md

The RightHub → Anaqua Journey

When I joined RightHub in 2023, it was a growing IP management platform with a clear vision: bring AI capabilities to the traditionally conservative world of intellectual property and legal tech. Within a year, this vision attracted Anaqua — a leader in IP management software — to acquire RightHub.

My role was to build the technical foundation that made this AI transformation possible.

Deep Dive: The AI Architecture

Vector Search Infrastructure

The core of our AI capabilities was a custom vector search system built on PGVector. Unlike generic implementations, ours was specifically designed for legal documents:

1
2
3
4
5

Document → Intelligent Chunking → Domain-Specific Embedding → PGVector Storage
                    ↓
            Citation Graph Building
                    ↓
Query → Embedding → Hybrid Search (Vector + BM25) → Re-ranking → Results

Key innovations:

  • Structure-aware chunking that keeps legal provisions, patent claims, and citations intact
  • Hybrid search combining semantic similarity with keyword matching for legal terminology
  • Citation-aware retrieval that follows reference chains to provide complete context

AI Agent Framework

Our agents weren’t simple chatbots — they were autonomous systems capable of:

  • Analyzing multi-page patent documents
  • Extracting and classifying entities (inventors, assignees, classifications)
  • Comparing documents and identifying conflicts
  • Generating summary reports with citations

The key to reliability was structured outputs:

1
2
3
4
5
6

class DocumentAnalysis(BaseModel):
    entities: List[Entity]
    classifications: List[IPClass]
    key_claims: List[Claim]
    confidence_score: float
    reasoning: str

Every agent action was validated against strict schemas, with graceful fallbacks and human escalation paths.

Technical Leadership

Beyond building systems, I established patterns and practices that the engineering team adopted:

  • AI Service Template: A standardized FastAPI service structure for all AI features
  • Prompt Engineering Guidelines: Documentation and examples for effective prompt design
  • Cost Monitoring Dashboard: Real-time visibility into LLM usage and costs
  • Testing Strategies: Evaluation frameworks for measuring AI output quality

The Acquisition

The AI capabilities we built were a key factor in the Anaqua acquisition. Our systems demonstrated that:

  • AI could be trusted with sensitive enterprise data
  • Complex legal workflows could be meaningfully accelerated
  • The architecture was scalable and maintainable for the long term

This validated my approach of building production-grade AI systems rather than impressive demos.

Technologies: LangChain, RAG Systems, PGVector, FastAPI, Spring Boot, OpenAI, Anthropic Claude, Python, AI Agents

Similar Roles: Senior Engineer at Flowrite | Founder at Sparrow Intelligence

$ ls -la case-studies/

Deep-dive technical case studies from this role:

📋 Enterprise RAG for Legal Documents - Vector Search at Scale → 📋 Multi-LLM Orchestration with Intelligent Cost Control →