timeline

Building core platform infrastructure for a multi-tenant agentic SOC product - agent harness, data service layer, observability, and enterprise auth.

Self-service analytics

• Built and owned a multi-tenant self-service analytics platform end to end, adopted across 100+ customers: Apache Doris, Postgres CDC with p95 lag under 30 seconds, query execution, and the dashboard UI. Rewrote the heaviest legacy aggregations for a 50x p95 improvement and moved that load off the Postgres primary.
• Made arbitrary customer SQL safe using RLS, curated views, column-level policies, and workload groups, resolved per query for each tenant and user. Customers can build queries, charts, dashboards, and alerts without reading outside their grant or starving another tenant's workloads.
• Built an LLM agent that authors widget SQL and runs cross-table investigations from natural language. Moved AI observability onto Doris, replacing Langfuse and ClickHouse, so traces, tool calls, token costs, and integration health are queryable and alertable in one place.
• Built the PostgreSQL-to-Doris CDC pipeline with monitoring and alerting for data freshness and pipeline failures.

Data service & integrations

• Built the vendor-agnostic, plug-and-play data layer that is the agent's hands and eyes: alert ingestion across 50+ integrations (CrowdStrike, Splunk, Palo Alto), processing ~50K alerts/day with 10–50 tools/alert, log search, enrichment, and response actions. Built these clients on one resilient integration layer for shared exception handling, retries, rate limiting, and later agent tool conversion, with customer-extensible MCP tools.
• Integrations are the critical path of every investigation, yet failures used to collapse into opaque errors. Surfaced first-class exceptions and a standardised error taxonomy shared across all integrations, so customers see exactly why something failed and oncall, notifications, and alerting run off one consistent signal.

Agent harness

• Worked on tool execution, permissioning, provider-specific prompt caching, and skills to enable customer-specific workflows. Reduced cost 10x and investigation time 3x with a ~95% prompt-cache hit rate.

Across the platform

• Enterprise auth & access control - Built multi-tenant AuthN/AuthZ for 10+ enterprise customers: SAML/Okta SSO, email OTP/password login, hierarchical RBAC with customisable roles, AuditLogs with sensitive-data redaction, database schemas and query optimisations, external API, and security-vulnerability fixes. Also shipped real-time i18n middleware (AWS Translate).
• Database reliability - Cut Postgres CPU load by 30% and improved p95/p99 latency ~20x by diagnosing a misordered composite index through production query-plan analysis.
• User comms & case management - Built case management across Teams, Jira, and Slack: auditable ticket-based cases, reminders, escalations, action triggers, and inline user context gathering.

Full-stack AI and large-scale automation across the enterprise.

• Built an end-to-end GenAI document-processing platform (classification, extraction, human review, translation) - onboarded 100+ businesses, processing 20K documents/day. Replaced Kofax/UiPath, saving $2M+ in operational and licensing costs. Core innovation: a meta-prompting system that consumes historic input/output pairs for a document type and auto-generates a structured extraction skill - converting vague, vendor-written rules across 45K+ document types into precise, example-grounded prompts without manual template work.
• Reduced new-document onboarding from ~3 months of bespoke Kofax UI, model training, and locator creation to providing representative documents. Improved extraction accuracy from 20% to 90%, with near-100% classification accuracy.
• Built an AI secondary-review workflow for stop-payment dispute processing, using RAG and a knowledge graph to replace manual QA and reduce resolution time and review cost.
• Built 5+ enterprise-grade AI/RPA solutions, saving $1M+ in operational cost, including LDAP-based entitlement validation across 100+ groups for 10K users/month. Recognised with Spotlight Awards in 2024 and 2025.
• Led the Records Retention Bot processing 3M+ files across 100+ Shared Drives and SharePoint sites - automated retrieval, archiving, and deletion via SharePoint APIs and Documentum.

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 for 5 EdTech companies - 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.