AI Engineer · Boston, MA · Open to roles

Models are confident. That isn't the same as right.

I build production LLM, agentic and retrieval systems — and the verification underneath them. The most useful thing I've learned: I once audited a model's confidence against its realized error and found the two uncorrelated. It was most certain exactly where it was wrong.

So I build gates. Deterministic verifiers that override the model. Evaluation harnesses that reject changes before they ship. Review queues ranked on measured error rather than what the model believes.

89 → 100%
metric coverage closed with calibrated uncertainty
shizen · production
87.8%
pass@1 on all 164 HumanEval+ problems
verified · csv + logs
$2.50 → $0.20
cost per unit after a reusable memory layer
shizen · production
5,457
review items cleared by folding fixes into rules
shizen · production

01

How I build

I'm an AI engineer who ships systems end to end — model, pipeline, service, and the evaluation that decides whether any of it is allowed into production.

At Shizen I was the sole engineer on an institutional risk platform: four Python services on Kubernetes, agentic document discovery across the open web, LLM extraction with page-level citation verification, and a bitemporal warehouse where every stored number traces back to the document and date it came from. No codebase, no team, first release in four months.

Before that, at Civica, I built distributed ETL on Databricks and Azure for financial and healthcare clients under regulated data-handling standards — 50K+ documents a day, 5+ TB, and the validation frameworks that kept malformed data out of downstream reporting.

The habit that runs through all of it: a wrong answer with no warning is worse than a missing one, because someone acts on it. So claims get checked against sources, changes get gated against held-out sets, and confidence gets audited rather than trusted.

I recently audited my own project repositories against their source and corrected several figures I'd been citing. The numbers on this page are the ones that survived.

At a glance

Focus
LLM systems · Agents · Retrieval · Evaluation
Education
MS Information Systems, Northeastern · Dec 2026
Based
Boston, MA · open to relocation
Domains
Financial risk · Healthcare · Supply chain · Public sector
Also
Reinforcement learning · Search · Statistical validation

02

Where I've built

Two roles: one startup from zero, one enterprise delivery environment.

Founding AI & Data Engineer

Shizen, Inc.
Jan 2026 – Apr 2026 · Boston, MA
  • Shipped an institutional risk platform as sole engineer with no codebase or team — four Python services on AWS EKS behind a REST API gateway, a bitemporal Postgres store, and a Next.js and React client — reaching first production release in four months.
  • Orchestrated agentic acquisition across ~33 heterogeneous sources, with planning agents navigating web archives, APIs and document repositories through multi-step tool use, clearing 1,100+ batches unattended at $0.20 per unit against a $2.50 baseline.
  • Enforced quality control on every generated record: page-level citation verification that overrode model output on conflict, plus unit-sanity, boundary and prior-year contradiction checks over a store with lineage and audit trails.
  • Trained and validated a random-forest predictor with conformal prediction intervals under a leakage-controlled masked backtest and temporal holdout, closing metric coverage from 89% to 100% under enforced release gates.
  • Audited model confidence against realized error, found the two uncorrelated, and re-ranked human review on measured error — surfacing 550 genuine issues and demoting 1,268 false flags.

Junior ETL Developer

Civica Resource Pvt. Ltd.
Jul 2022 – Jul 2024 · India
  • Delivered distributed PySpark pipelines on Databricks for financial and healthcare clients, processing 50K+ documents daily into a layer teams could query same-day.
  • Automated scheduled ETL in Azure Data Factory across 5+ TB and 3M+ records, deploying through Azure DevOps with real-time metrics monitoring.
  • Hardened SQL Server and PL/SQL pipelines with automated validation frameworks, intercepting malformed upstream data at ingestion.
  • Mentored teammates onto Databricks workflows and codified pipeline standards — C-STAR Award for leadership on data pipeline design.

03

Projects

Reinforcement learning, search, generative modeling and quantitative risk — built from papers, run on real hardware, and audited against their own logs.

Reinforcement Learning

Policy optimization from scratch

Three distinct algorithms implemented against Qwen2.5-3B with no reference code: PPO with GAE and a learned critic, an advantage-weighted variant using a smooth-max reference-model baseline with KL regularization, and REINFORCE. Includes V* caching, curriculum learning, and a variance-based trajectory filter.

88%
countdown + multiplication
39+
H100 runs on Modal
PyTorchQwen2.5-3BPPO / GAEModal H100
Agentic Systems

Learned agent orchestration

A REINFORCE controller trained to select which reasoning, refinement and validation operators to invoke per problem, optimizing score minus weighted cost. Trained end to end with a checkpoint the evaluation provably loaded. Also surfaced a code-extraction defect firing on 44% of intermediate test-operator executions.

87.8%
pass@1, 144/164
44%
defect rate found
REINFORCEPolicy gradientGPT-4o-miniHumanEval+
Quantitative Risk

Derivatives risk engine

An end-to-end risk analytics system: arbitrage-free SVI volatility surfaces calibrated under Durrleman butterfly and calendar constraints, VaR and Expected Shortfall computed by full revaluation and Taylor expansion in parallel, Kupiec and Christoffersen backtests, PCA risk factors, and minimum-Mahalanobis reverse stress.

VaR / ES
two independent methods
PCA
factor decomposition
NumPy / SciPyDuckDBNumbapytest
Search · Planning

MCTS and LLM-guided search

Monte Carlo Tree Search with UCT implemented from scratch — selection, expansion, simulation, backpropagation — then extended with language-model priors to solve the 24-Game. Solvability gating and integer bias lifted accuracy well above both the heuristic baseline and value-only LLM guidance.

1.00
grid success at 200+ sims
80%
24-Game vs 60% baseline
MCTS-UCTGPT-4oPython
Retrieval · Full-stack

Filing intelligence assistant

A retrieval pipeline over SEC 10-K and 8-K filings that embeds semantically chunked passages into a FAISS index and returns the source passage behind every answer. Shipped as a self-serve application so non-technical users could pull structured records without an engineer in the loop.

FAISS
vector retrieval
Live
deployed on EC2
LangChainOpenAIFAISSStreamlit
Generative Models · Privacy

Identity anonymization with GANs

Conditional identity-anonymization GANs reimplemented in PyTorch, training adversarial generators that strip identifying features from personal images while preserving the attributes downstream models depend on, with a disentanglement loop separating identity from non-identity representation.

GAN
adversarial training
DisCo
disentanglement
PyTorchConditional GANsComputer vision
Foundations

Multi-armed bandits

ε-greedy and gradient bandit algorithms implemented from scratch and evaluated across 2,000 runs of 1,000 steps, reproducing Figure 2.2 from Sutton & Barto and comparing softmax preference learning with and without a baseline at two learning rates.

2,000
runs × 1,000 steps
Fig 2.2
reproduced
ε-greedyGradient banditsNumPy
Production Platform

Shizen data engine & warehouse

The platform behind my last role: agentic document discovery and LLM extraction feeding a bitemporal medallion warehouse on Postgres, with Kimball dimensional models, point-in-time-correct fact tables, and a governance layer for lineage, audit and validation.

4
Python services
~33
ingestion sources
AWS EKSVertex AIAirflowPostgres

04

Toolbox

What I reach for, grouped by the problem it solves.

Languages

Python · SQL · TypeScript / JavaScript · Java · Shell

LLM & Agentic Systems

Prompt engineering · Structured output · Function calling · Multi-agent orchestration · Model routing & cost control · LangChain · LangGraph · MCP · Claude, GPT, Gemini

Retrieval & Memory

RAG · Embeddings · Vector search (FAISS, Pinecone) · Semantic chunking · Document AI / OCR · Run & session memory

Evaluation & Verification

Eval harnesses & A/B gating · Hand-labeled eval sets · Deterministic verifiers · Domain guardrails · Calibration auditing · Release gating · Failure & ablation analysis

Machine Learning

PyTorch · scikit-learn · Hugging Face · Transformers · Model fine-tuning · Reinforcement learning (PPO, A*-PO, REINFORCE) · Multi-armed bandits · Monte Carlo methods · Conformal prediction · GANs

Data Engineering

PySpark · Databricks · Airflow · dbt · Azure Data Factory · PostgreSQL · Redis · DuckDB · Snowflake · Bitemporal modeling · Lineage & audit

Backend & Infrastructure

FastAPI · REST API design · Microservices · Docker · Kubernetes (EKS) · Event-driven architecture · Dead-letter queues · Circuit breakers · Rate limiting

Cloud & DevOps

AWS (EKS, EC2) · GCP (Vertex AI, Cloud Run, Document AI) · Azure · Modal (serverless GPU) · CI/CD (GitHub Actions, Azure DevOps) · Git

Quantitative

Statistical modeling · Leakage-controlled backtesting · Temporal validation · Time-series analysis · PCA · Constrained optimization · VaR & Expected Shortfall · Coverage tests (Kupiec, Christoffersen)


05

Academics

M.S. Information Systems

Northeastern University · Boston, MA
Expected Dec 2026
Self-Improving AI · High Performance ML & AI · Data Science Engineering Methods & Tools · Program Structures & Algorithms

B.Tech. Industrial & Production Engineering

Jabalpur Engineering College · India
May 2022
Operations Research · Operations Management · Material Management & Product Design · Industrial Robotics & Mechatronics
06

Let's talk

Open to AI and ML engineering roles. Email is fastest — I reply the same day.