Tirth Patel
Software engineer, AI systems and backend. Toronto, Ontario. Eligible to work in Canada.
- ptirth339@gmail.com
- Web
- tirthpatel.codes · github.com/TirthPatel6104
- linkedin.com/in/tirthpatel610
BSc Computer Science (Trent University, December 2026, GPA 4.0) who ships tested, measured systems. Built a full-stack forestry growth-modelling platform on real provincial data for scientists at Ontario's Ministry of Natural Resources and Forestry, with a 17-algorithm leakage-aware benchmark. Author of two open-source AI infrastructure tools: ContextBridge and Substrate.
Experience
Software Developer, Industry Partner Project
Client: Ontario Ministry of Natural Resources and Forestry (MNRF).
- Built the Forest Growth and Yield Analytics System (Python, FastAPI, SQLAlchemy 2.0, Alembic, React 19, TypeScript, Vite) that fits, benchmarks, and serves stand-level basal-area growth models for 13 species and forest-unit groups on real MNRF inventory data.
- Implemented the production relative-growth-rate regression engine with biological sign-plausibility checks on every fitted coefficient, exposing per-plot predictions and 5 to 25 year projections through 25 REST endpoints.
- Designed a leakage-aware benchmark of 17 algorithms (OLS, ridge, lasso, spline GAM, SVR, random forest, XGBoost, LightGBM, MLP, mixed-effects) under plot-grouped, temporal, and naive cross-validation, with paired t-tests against the production equation, split-conformal prediction intervals, and
permutation_importance; published a JSON and Markdown leaderboard across all 13 groups. - Engineered a portable data layer (SQLite, PostgreSQL, SQL Server behind one Alembic chain) and asynchronous training and benchmark jobs (202 Accepted and polling) so long fits never block the API.
- Set up CI quality gates: 89 pytest tests with coverage, flake8 and mypy, ESLint and a TypeScript build, Playwright end-to-end tests with axe accessibility scans, a migration sanity check, and a benchmark smoke test on GitHub Actions.
- Worked directly with MNRF forestry scientists as clients, turning growth-and-yield requirements into specifications; wrote C4 system design, benchmarking, and security-disclosure documentation.
Projects
ContextBridge: portable, privacy-conscious memory for LLMs
Python 3.11+, Pydantic 2, FAISS, SQLite, Flask, Click, Chrome extension (MV3), GitHub Actions
- Extracts memory from ChatGPT and Claude conversations into six typed categories, redacts secrets and PII before storage, and renders model-specific prompts, with versioned packages, rollback, and conflict-flagged merging. One service layer serves a CLI, a local dashboard, and a Chrome extension. 201 offline tests, ruff, CI on Python 3.11 to 3.13.
- Explainable hybrid retrieval (BM25 with optional embedding blending) that reports matched terms, score, and reason per item; cuts injected context by 93.2% against the raw transcript and 81.5% against full memory.
- Offline evaluation suite (cb eval): precision@3 76.7%, recall@3 93.3%, MRR 86.7%, duplicate-detection F1 75.0%, redaction recall 100% with 0 false positives. Regression floors asserted in CI.
Substrate: local-first intelligence layer for AI agents
Python (standard library only), SQLite FTS5, Model Context Protocol, JSON-RPC over stdio
- MCP server with 17 tools that give any coding agent persistent typed memory with provenance, incremental file indexing with hybrid FTS5 and vector recall, and a durable task ledger with lease-based locking for multi-agent handoff. Zero network calls by default.
- Deterministic, LLM-free command-safety gate (SAFE, NEEDS_CONFIRMATION, HARD_BLOCK) with a persistent approvals queue and audit log; regression tests pin known bypasses shut. 58 tests, CI on Python 3.10 to 3.12.
Privacy Bridge
Next.js 16, React 19, TypeScript, Solana Web3.js, Supabase
- Hackathon MVP of a cross-chain bridge that executes real Solana wallet transactions (Phantom, Solflare, Backpack) and reconciles them against a Supabase transaction ledger with Explorer-linked history.
Skills
- Languages
- Python, TypeScript, JavaScript, SQL, Java, C#, Bash
- Backend and APIs
- FastAPI, Flask, Pydantic 2, SQLAlchemy 2.0, Alembic, REST API design, OpenAPI, background jobs, JSON-RPC
- AI and LLM
- RAG, BM25 and embedding retrieval, FAISS, LLM integration (OpenAI, Anthropic, Ollama), Model Context Protocol, prompt engineering, LLM evaluation (precision@k, MRR), PII redaction
- ML and data
- scikit-learn, XGBoost, LightGBM, statsmodels, pandas, NumPy, grouped cross-validation, conformal prediction
- Front end and data stores
- React 19, Next.js, Vite, Tailwind CSS, Radix UI, Recharts; PostgreSQL, SQLite (FTS5), SQL Server, Supabase
- DevOps and practice
- Git, GitHub Actions, Docker, pytest, ruff, mypy, ESLint, Playwright, axe; code review, TDD, Agile
Education
BSc Computer Science, Trent University
Peterborough, Ontario. GPA 4.0 of 4.0.