Skip to content

01AI/ML · Data Engineering · Full Stack

AEGIS AI

Real-time fraud detection and autonomous investigation

A streaming fraud platform that scores card transactions as they happen and explains every alert, built on synthetic data.

  • Redpanda (Kafka API)
  • PostgreSQL 16
  • Redis 7
  • scikit-learn
  • pandas
F1, gradient boosting (test split)
0.863
F1, gradient boosting (test split)
PR-AUC on 2.18% fraud prevalence
0.926
PR-AUC on 2.18% fraud prevalence
alert rate on normal traffic
0.30%
alert rate on normal traffic

02AI/ML · Data Engineering · Full Stack

NEXUS AI

Multi-agent research and execution workspace

A local-first AI workspace where every answer keeps its plan, its agents, its sources and the SQL behind each number.

  • LangGraph
  • Ollama (qwen3:4b-instruct)
  • PostgreSQL + pgvector
  • DuckDB
  • Redis
citation grounding checks
6/6
citation grounding checks
unsafe SQL statements blocked
8/8
unsafe SQL statements blocked
canary leaks
0/5
canary leaks

03AI/ML · Full Stack

SourceAtlas

Agentic RAG with verified citations

A research engine that checks every claim against the passage it cites, surfaces conflicts between documents and abstains when it can't verify.

  • LangGraph
  • PostgreSQL 16 + pgvector
  • Ollama (qwen3:4b-instruct)
  • Python 3.13
  • FastAPI
answer correctness, live eval (16 questions)
92.3%
answer correctness, live eval (16 questions)
citation correctness, live eval
95.8%
citation correctness, live eval
Recall@5
1.00
Recall@5

04AI/ML · Full Stack

TuneLab

LoRA fine-tuning with honest evaluation

Trains LoRA adapters for a small LLM on Apple Silicon and proves what changed by evaluating against the frozen base model on identical prompts.

  • FastAPI
  • SQLite (WAL)
  • Next.js 16
  • React 19
  • Tailwind 4
all-fields exact match, test split
8.3% → 100%
all-fields exact match, test split
on held-out phrasings
5.0% → 63.3%
on held-out phrasings
of parameters trained (2.93M of 494M)
0.59%
of parameters trained (2.93M of 494M)

05AI/ML · Data Engineering · Full Stack

InsightForge

An autonomous data analyst with sandboxed SQL

Ask a question of your data; the LLM plans, typed operations and sandboxed SQL execute, and every number in the explanation is verified.

  • DuckDB
  • pandas
  • scikit-learn
  • LangGraph
  • Ollama (qwen3:4b-instruct)
unsafe-SQL rejection and injection safety
100%
unsafe-SQL rejection and injection safety
numerical correctness with a local 4B model
66.7%
numerical correctness with a local 4B model
layers of SQL isolation
4
layers of SQL isolation

06AI/ML · Full Stack

ReviewPilot

Multi-tenant GenAI code-review SaaS

Teams submit diffs; deterministic rules and a local LLM review them, and every model finding must quote the lines it cites or it is discarded.

  • NestJS
  • PostgreSQL 16
  • Redis 7
  • Ollama (qwen3:4b-instruct)
  • Next.js 16
deterministic review rules
17
deterministic review rules
tenant roles with server-side enforcement
4
tenant roles with server-side enforcement
automated tests across layers
123
automated tests across layers

07Full Stack

WorkMesh

Permission-aware company visibility

Connects org structure to the work happening inside it, so people can see who is doing what, within the permissions they actually have.

  • NestJS 11
  • PostgreSQL 16 (RLS, FTS)
  • Redis 7
  • Next.js 16
  • React 19
p50, employee dashboard (300 requests, concurrency 10)
80 ms
p50, employee dashboard (300 requests, concurrency 10)
errors across the latency benchmark
0
errors across the latency benchmark
tenant tables under row-level security
35
tenant tables under row-level security