Track B · Parallel portfolio · Internship-oriented
Quant developer roadmap
Become competitive for quant developer / trading systems / market infrastructure internships —
not quant researcher roles. Leverage your Go, distributed systems, and crypto-payments background.
Structure mirrors roadmap.sh’s progressive style, with nested math and AI-engineering branches for how trading firms actually work in 2026.
LOB · matching · data
Math fluency via code
AI-assisted quant eng
What “LOB” means here
Limit Order Book — the live book of buy/sell limit orders an exchange (or your simulator) matches against.
Building a LOB + matching engine is the highest-signal personal project for a systems engineer entering quant-dev.
How AI changed quant engineering
What changed
- Feature pipelines & ML models in production alpha/risk systems
- LLMs for research notes, code, ops runbooks — with hallucination risk
- AI-assisted coding raises the bar for verification and testing
- Crypto/DeFi still needs deterministic engines; AI doesn’t remove microstructure
What did not change
- Correct order lifecycle, PnL, and risk limits still win interviews
- Latency, determinism, replayability still matter
- Firms still grill you on probability, regressions, and edge cases
- You are not hired to “prompt GPT to trade” — you build systems
Progressive path
Ship Q1 + Q2 before expanding. Cyber remains your degree spine; this is the parallel portfolio.
0 · Quant developer vs quant researcher
Mindset
- Developer: engines, data, execution, risk infra, correctness, performance
- Researcher: signals, papers, novel models — not your primary target
- Resume language: “Trading Systems / Market Infrastructure”, not “Quant Research Intern”
- Success metric: repos that interviewers can clone and run tests on
1 · Market microstructure vocabulary
Required
- Orders: market, limit, stop; TIF (GTC/IOC/FOK)
- Bid/ask, mid, spread, depth, imbalance
- Fees, rebates, maker/taker
- Slippage, market impact (intuition)
- Exchange vs broker vs OTC; crypto venue quirks (partial books, disconnects)
Resources
- BookLarry Harris — Trading and Exchanges (order & market chapters)
- DocsBinance / Bybit / coinbase exchange API docs — read order types carefully
- EssayClassic HFT/microstructure explainers (filter hype; keep definitions)
2 · Limit order book & matching engine
Flagship eng
- Price-time priority book data structures
- Add / cancel / amend / match; partial fills
- Self-trade prevention; reject paths
- Deterministic replay from an event log
- Benchmarks: throughput (orders/sec), latency histograms
- Concurrency: single-writer vs sharded books; correctness first
Resources
- ImplYour Go concurrency / profiling skills — pprof the hot path
- RefOpen-source matching engines — read, don’t copy blindly; cite ideas
- TestProperty-based tests for conservation of quantity
3 · Market data systems
Required
- Ingest trades, BBO, depth via WebSocket
- Normalize schema; handle gaps, dupes, out-of-order events
- Store Parquet / TimeSeries DB; partition by date/symbol
- Rebuild book state at time T from the log
- Clock issues: exchange vs local time; monotonic sequencing
Resources
- ToolApache Arrow / Parquet; Timescale or ClickHouse (pick one)
- BookLópez de Prado — Advances in Financial ML (data hygiene chapters only)
4 · Execution sim, backtest & risk controls
Recommended
- Order lifecycle in the simulator (submitted → ack → partial → filled)
- Fees, slippage, latency models — make assumptions explicit
- Mark-to-market PnL, equity curve
- Kill switch, max loss, position limits, notional caps
- Optional: TWAP/VWAP-style parent/child slicing
- Strategies stay thin — the engine is the product
Resources
- PracticeBacktest “gotchas” checklists (look-ahead bias, survivorship)
- RefExchange fee schedules — encode real numbers in tests
5 · Math fluency (nested skill track)
Nested
Required
Prove fluency by implementing + verifying — not by collecting courses.
- Probability: expectation, variance, distributions, LLN/CLT intuition
- Returns: log vs simple; volatility; Sharpe (compute carefully)
- Dependence: correlation/covariance; why they break; PCA on returns
- Regression: market beta, residuals, multicollinearity intuition
- Stochastic lite: GBM paths; Black–Scholes as verification target
- Monte Carlo: error vs √N, CIs, antithetic variates
Resources
- BookBlitzstein — Introduction to Probability (free)
- BookHull — Options, Futures… (BS + Greeks chapters)
- BookGlasserman — Monte Carlo Methods (skim + implement basics)
- Video3Blue1Brown — Essence of Linear Algebra
6 · Numerics, money & testing discipline
Required
- Decimals/integers for money; floats for sims — know when each breaks
- Reproducibility: seeds, golden files, deterministic event ordering
- Property tests: conservation of shares, non-negative queues
- Benchmark honestly (warm-up, distributions, not single means)
- Write MATH.md / DESIGN.md that answer “how do you know it’s right?”
7 · AI in quant engineering (nested: AI Engineer lite)
AI-era
Nested
Borrow from roadmap.sh/ai-engineer:
apply models/tools — don’t train foundation models.
- ML features in production: offline/online skew, leakage, monitoring drift
- Classical ML enough: linear/logistic, trees, simple neural nets for features — sklearn level
- LLM tooling: use for code/docs; require tests; never trust generated PnL logic untested
- RAG over research: optional internal wiki search — watch prompt injection if tools can trade (don’t let them)
- Eval mindset: backtests and paper metrics beat vibes — same as LLM evals
- Security crossover: protect keys, trading APIs, research data — your cyber track helps
Resources
- Roadmaproadmap.sh/ai-engineer — application focus
- BookHands-On ML (Géron) — selective chapters
- PracticeBuild a feature + naive model on your Parquet store; measure IC/PnL carefully
8 · Languages & stack choices
Leverage Go
- Go: matching engine, ingest, services — your strength (ship Q1 here)
- Python: research, MC, stats, glue, notebooks → scripts
- SQL: analytics over stored ticks
- C: useful literacy only (same as cyber) — not a quant-dev priority
- C++: valuable later if targeting classic prop / HFT-style shops; not required for crypto/fintech/market-infra internships if Go+Python portfolio is strong
- Rust: optional alternative to Go/C++ for engines — only if curious
When to actually learn C++
Now: don’t. Finish LOB (Go) + Monte Carlo (Python) first.
Later: if interview targets list C++ (many London/NY prop firms, some EU market-makers), add a focused slice —
modern C++ (RAII, move semantics, STL), concurrency, and rebuild a slice of your matcher in C++.
Blindly studying C++ before a portfolio exists slows both tracks.
9 · Interview readiness
Later stage
- Explain your LOB design and failure modes on a whiteboard
- “How do you know your Monte Carlo price is right?”
- Coding: arrays, heaps, concurrency bugs, parsing streams
- Stats: log returns, correlation pitfalls, overfitting in backtests
- Target list: crypto trading firms, market-infra startups, fintech execution teams in France/EU
Projects
Q1 · Systems flagship
LOB + matching engine
- Go engine + tests + benchmarks
- Deterministic replay
- 3–5 min demo video
Q2 · Math flagship
Monte Carlo ↔ Black–Scholes
- Verification tables + error plots
- Variance reduction
- MATH.md with assumptions
Q3 · Data + execution
Ingest + backtester
- Parquet store + replay
- Fees/slippage/latency model
- Thin sample strategy only
Q4 · Optional
Returns / PCA / beta toolkit
- Or small ML feature experiment with leak checks
- Only after Q1 + Q2 are solid
Minimum viable portfolio
Q1 + Q2 with crisp READMEs. That pairing proves systems competence and math fluency — enough to apply for quant-dev shaped internships while your cyber degree runs.
Resource library
| Area | Primary | Secondary |
|---|
| Microstructure |
Harris — Trading and Exchanges |
Exchange API docs |
| Probability |
Blitzstein Intro to Probability |
Worked problems in code |
| Derivatives math |
Hull (BS + Greeks) |
Your MC verification project |
| Monte Carlo |
Glasserman (selective) |
NumPy implementations |
| Linear algebra |
3Blue1Brown |
PCA-from-scratch once |
| Data hygiene |
AFML (selective) |
Parquet + replay tests |
| AI application |
roadmap.sh/ai-engineer |
Géron Hands-On ML (selective) |
| Systems perf |
Go pprof / benchstat |
Your production instincts |