timeline

Building the integrations platform and platform-wide reliability for a multi-tenant agentic security product - a SOC-investigation agent that triages alerts end-to-end.

• Agent tool surface - the integrations platform is the agent's hands. Plugged in third-party security vendors (SIEM, ticketing, comms) as alert/incident ingestion, enrichment, response actions, and custom playbook hooks. Shipped 20+ integrations, 50+ security tools on this surface.
• Agent tool-use, debugged and extended - when the SOC agent drifts on a real alert - wrong tool picked, wrong params - I dig into the trace and fix it at the tool definition, argument schema, or prompt level. Also packaging new vendor actions into GRPC/MCP tools for the agent.
• Backend API & data model - owned the core Django ORM models, REST surface, and auth/permissions around vendor connection lifecycle, action invocation, and audit + retry state. Turning ambiguous customer asks to platform features within days.
• Unified integration observability - normalised API error logs that were scattered across vendor-specific paths into a single structured event shape. Built the cross-integration dashboard for P95/P99 latency, error rate, and throughput, plus alerting on failure rates and a single health view across all integrations.
• Database reliability - cut production Postgres CPU 100%→60% and API latency p95 ~20x by fixing an incorrectly ordered composite index surfaced through query-plan analysis in prod; diagnosed connection-pool exhaustion under burst traffic and validated the PgBouncer migration.
• User-in-the-loop comms - built the Microsoft Teams service that drives investigation lifecycles for end users: notifications, structured input prompts, case reminders, escalations, and context gathering inline in chat.

Backend GenAI and large-scale automation across the enterprise.

• Built a Retrieval-Augmented Generation (RAG) automation for transaction stop-payment dispute processing using LangGraph, Neo4j knowledge graphs, and OpenAI APIs - improving accuracy and cutting resolution time.
• Developed a generic, fully scalable agentic AI solution for document data extraction across 45K+ document types.
• Built 5+ enterprise RPA solutions in UiPath, Python, and .NET - reduced manual effort and saved $1M+ in operational cost. Recognised with Spotlight Award (2024 & 2025).
• Led the Records Retention Bot processing 3M+ files across 100+ SharePoint sites - automated retrieval, archiving, and deletion via SharePoint APIs and Documentum.
• Building a full-stack intelligent document processing platform (classification, extraction, translation) in Python, MongoDB, OpenAI, and React - designed to onboard 100+ businesses.

Applied research lab building self-programming machines - natural language → custom software via a flow-based DSL, drawing on the DreamCoder line of work on wake-sleep bootstrapped library learning.

• Flatland augmentations - Flatland is the team's toy DSL and benchmark for synthesising 2D line-drawing programs: a system that learns to draw straight lines and circles, then after each round of training on augmentations of its own programs, learns squares and rectangles, then richer compositions. Built the augmentation generator that produced the training corpora each round bootstrapped on.
• Platform connectors - first-class integrations for Notion, Google Sheets, Gmail - the surfaces end-users wired their natural-language programs to.
• HuggingFace integration - added the HuggingFace model registry to the platform so fine-tuned LMs could be swapped into the program-synthesis pipeline.

Built and deployed math OCR models for an EdTech platform.

• Developed MathOCR and GeometryOCR for converting printed/handwritten math images to LaTeX; built a 10M+ image dataset using NumPy, Matplotlib, and OpenCV with custom augmentation pipelines.
• Optimised models with TensorRT and deployed via Triton Server - sustained 5000 req/min on 1× NVIDIA T4.
• Wired Prometheus + Grafana for inference metrics and alerting.
• Used GCP and AWS for cloud storage, model training, deployment, and inference at scale.

Embedded AI for a real-time mask, temperature, and face recognition system.

• Developed and deployed on Raspberry Pi and Jetson Nano; generated a 100K+ image dataset from augmented scraped images.
• Converted the model to TensorRT - achieving 50ms inference latency on edge hardware.
• Used PyQT for the operator UI, MySQL for data storage, and Grafana for logging.