Projects

AI equity research reports — grounded in SEC filings

A founder project: an AI equity research platform that turns SEC filings, earnings-call transcripts and public company documents into cited, verifiable research reports for individual investors-no advisory credentials or institutional access required.

Two modes for different needs: a ~20-second Snapshot for fast screening, and a ~3-minute Full Report covering business model, competitive moat, management tone, earnings-call Q&A and risks beyond the financials. Additional tooling includes Ticker Compare, monthly portfolio auto-recaps and quarter-over-quarter report diffs.

Engineered for trust over guesswork: SEC EDGAR ingestion (10-K, 10-Q, 8-K), earnings transcripts via Azure Document Intelligence, market data with Yahoo Finance/Finnhub failover, and parallel multi-agent drafting (Claude, Gemini, GPT) with reconciliation and citation anchoring-every material claim links back to its source document.

The AI Engine for Enterprise-Grade Process Automation

A multi-tenant SaaS platform for AI process automation: knowledge-aware assistants, document workflows, and deep integration into real back-office operations (not demo workflows).

Built on AWS with event-driven microservices (Lambda, ECS/Fargate, SNS/SQS) and multi-model LLM integrations (OpenAI, Azure OpenAI, Bedrock, Mistral). Designed for high-throughput workloads (>1,000 concurrent connections; thousands of requests per second) with controlled cost and reliability.

Enterprise-grade security and compliance by design: ISO 27001 and SOC 2, with data sovereignty as a first-class requirement.

Enterprise tooling for reliable LLM pipelines

Built an internal testing and evaluation framework (Testbed) to make LLM systems measurable and repeatable: deterministic test cases, quantitative + qualitative evaluation, and regression testing for RAG and multi-step pipelines.

Developed "Prompt Wizard" to streamline prompt iteration and optimization-helping teams ship prompts that are robust, versioned, and testable instead of "works on my laptop".

The goal: move prompt engineering from ad-hoc experimentation to engineering discipline-so teams can scale, ship safely, and keep performance stable over time.

Automated product matching to eliminate manual offer preparation

For RP Group (emergency lighting), built a machine-learning system that matches customer inquiries to the best-fitting products-replacing time-consuming expert-driven selection with repeatable ranking.

Built a robust dataset via LLM-assisted extraction (entAIngine API) and domain-specific feature engineering (25+ engineered features). Benchmarked five models with hyperparameter optimization.

Best model: tuned Random Forest with 70.7% Top‑1 accuracy (MRR 0.835; AUC‑ROC 0.923), delivering strong ranking quality while keeping the system interpretable and deployable.

Save 2+ hours per appraisal with AI-driven research

An AI platform built for real estate appraisers and valuers to automate the most time-consuming parts of an appraisal: research, location analysis, and report drafting.

Turns hours of manual work into 2–3 minutes of structured output per property with macro/micro location insights, market context, and appraisal-ready report generation.

Designed for professional workflows: multi-source data integration and quality controls to keep outputs consistent and reviewable.

Build multi-agent systems without writing code

A no-code platform enabling operations teams (customer service, HR, internal ops) to design and run multi-agent workflows without engineering support.

Focused on orchestration and reliability: clear role separation, controlled handoffs, and evaluation hooks so teams can iterate safely and scale outcomes across departments.

Due diligence and portfolio intelligence-productized

A VC-focused platform combining deep research with investment-specific automation to speed up due diligence and market intelligence.

Pre-configured agents support competitive analysis, portfolio intelligence tracking, deal flow automation, and market mapping-turning scattered research into institutional knowledge.

Safe Data Collaboration Without Exposing Sensitive Information

An EU Horizon 2020-funded platform enabling organizations to run analytics on private data without taking custody of it-"bring the code to the data" with strong auditability and security.

Designed end-to-end secure workflows (agreement → execution → evaluation) and a modular microservice architecture, plus immutable audit trails via blockchain.

Used confidential computing (TEEs / Intel SGX) and encryption for data at rest to support privacy-preserving, inter-organizational processing-validated with EU stakeholders.